fable is a local memory and recall server for Claude Code that indexes your entire conversation history, making it searchable and retrievable mid-session — operating entirely locally without API keys or cloud services.
fable_search – Search across all past sessions by keyword, project, tag, file, or action verb; returns ranked threads with metadata (IDs, token counts, decisions).
fable_thread – Retrieve exact, verbatim turns from a past conversation thread under a token budget — not a summary, the actual transcript.
fable_block – Recover a specific tool result or turn byte-for-byte by UUID.
fable_prune – Strip tool noise, images, and bloat from the current session transcript, with the original automatically sealed in a vault backup.
fable_remember – Store durable facts, preferences, or rules that are auto-injected at the start of every future session (optionally scoped to a project).
fable_recall – Read back all previously stored durable facts, globally or per-project.
fable_context – Auto-assemble a budget-aware context pack from the most relevant past threads for a given task.
fable_files – List all files Claude has edited across the archive or within a session, with edit counts and timestamps.
fable_file_history – Reconstruct every version of a file Claude has ever edited, with timestamps and session info.
fable_file_diff – Generate a unified diff between any two reconstructed historical versions of a file.
fable_tags – Discover taxonomy tag families and values for precise tag-filtered recall searches.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@fable MCP server what did we discuss about the database schema?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Quickstart · Why fable · Benchmarks · Roadmap · Discussions
Every conversation you've ever had with Claude Code is already saved on your machine — every decision, every debugging hunt, every 2 AM breakthrough, word for word. Claude just isn't allowed to use it.
Mid-session, compaction builds a wall: everything behind it is locked away
to save tokens, and Claude carries on with a thin summary. /clear wipes
the slate. And after 30 days, Claude Code quietly deletes the files
themselves. Your project's real memory — locked, then destroyed, by
design.
Everyone else sells you a replacement memory: summaries, extracted facts, vector stores. fable does something different — it unlocks the real one.
pipx install git+https://github.com/grooverLab/fable
fable install # one command: register the MCP, install hooks, index your history
fable serve # browse your memory in a dashboard100% local · no API keys · no cloud · no daemons. Your conversations never leave your machine.
pipx install git+https://github.com/grooverLab/fable # install the CLI
fable install # one-shot: claude mcp add + Claude Code hooks +
# ~/.fable home + index every transcript
fable serve # dashboard at http://127.0.0.1:8765
pipx upgrade fable-recall # update later (pulls latest from GitHub)fable install is idempotent — safe to re-run; it skips anything already wired.
To do it by hand: fable setup (home) · claude mcp add fable -- fable mcp (MCP) · fable discover (index).
command | what it does |
| rank threads by relevance ( |
| assemble a budgeted context pack |
| a thread's raw turns, byte-identical |
| a file's full edit history across sessions |
| generate AI summary cards (background) |
| (re)scan + index all Claude Code projects |
| slim a session · export · index stats |
| the dashboard |
Ask about past conversations with Claude — get the real answer. "What did we decide about auth last month?" Claude searches its own history mid-session (via MCP) and quotes the actual transcript — not a summary, not an extracted "fact." The conversation itself.
The wall stops costing you. fable catalogs everything before compaction walls it off, and hands back exactly what Claude lost — on demand, under a token budget you set.
The 30-day deletion becomes irrelevant. Sealed into a local vault, byte-identical, for as long as you decide.
Related MCP server: acheron-mcp-server
🧵 Composed Sessions. Hand-pick conversations from any project, any month — put them in your order — and fable builds a brand-new session that Claude resumes as its own lived history. A workspace with curated memory. (Empirically verified: restitched sessions resume cleanly, signatures intact.)
🕰️ File time-travel. Your transcripts accidentally versioned
everything. fable reconstructs every file's edit history — every Edit
and Write Claude ever made, across every session — with side-by-side
comparison between any two moments of a file's life, and a jump back to
the conversation that made each change. (fable file src/loader.py)
✂️ Transcript Surgery. Your 80 MB session is paying rent on dead threads. Remove whole conversations — fable re-stitches the timeline, shows you the simulation first, and keeps every removed byte recallable forever. Reversible by construction.
🪶 Pruning that loses nothing. Slim every message (tool noise, images, bloat) before resuming a heavy session — with an itemized preview of the savings, and the original sealed in the vault first.
🔍 Memory Diff. See exactly what any prune or cleanup cost any conversation — generation by generation, byte by byte. Nobody else can even show you what was lost.
🔇 Run cold. Mute recall for a session or a fleet of sub-agents and they
work with no prior context at all - no injected memory, no recall tools - while
capture keeps running, so the muted work is still recorded verbatim and
searchable the moment you lift it. The point is an unbiased second opinion:
an audit that cannot be anchored to yesterday's conclusion.
(fable mute on · FABLE_MUTE=1 claude · a toggle in Settings)
fable indexes your transcripts into a local SQLite archive: an immutable vault (every byte, forever) plus a search map (keyword + semantic, optional local embeddings via Ollama). Hooks run before Claude Code's compaction and cleanup; an MCP server gives Claude search / recall / remember tools. Measured on a real archive — 191,000 records, 6,000 conversations:
recall@1 | recall@5 | search (p50) | full re-index |
76.7% | 90.0% | 135 ms | 6.6 s |
Reproduce it: python3 scripts/benchmark.py. No competitor publishes
retrieval numbers.
fable | claude-mem | mem0 / Letta | native Claude Code | |
Memory = your actual conversations | ✅ | ❌ summaries | ❌ fact snippets | ⚠️ locked behind the wall |
Survives the 30-day deletion & | ✅ | ⚠️ its summaries do | ❌ | ❌ |
Claude searches its own history (MCP) | ✅ | ✅ | ❌ | ❌ |
Composed sessions / file time-travel / surgery / diff | ✅ first of its kind | ❌ | ❌ | ❌ |
Zero API keys, fully offline | ✅ | ✅ | ❌ | ✅ |
Footprint | one SQLite file | Node + Chroma daemon | cloud / Docker | — |
(Fair is fair: mem0 fits multi-LLM production agents; ccusage goes deeper on billing analytics. Different jobs, both compatible with fable.)
fable's first user is the session that built it: mid-build, that session
was pruned by fable (7.8 MB → 3.0 MB), kept working through compaction via
its own hook, and is now searchable through its own MCP server — and
fable file fable/recall.py replays its own source code being written,
18 versions deep. The build history eats its own dogfood — all $83k of
API-equivalent work in the author's archive included.
What people use it for
"Why did we choose X?" — architecture archaeology, weeks later
"When did this function break?" — file time-travel to the exact edit and the conversation around it
Picking up a debugging hunt exactly where the wall cut it off
A composed "workspace" session: threads from three projects, one memory
Slimming a heavy session before --resume, reversibly
fable remember "we deploy Fridays only" — standing rules, every session
Try it on a fictional sample first:
python3 demo/seed_demo.py && fable --db demo/demo.db serve
MIT · local-first forever (non-goals) · built with Claude
Code, for Claude Code · @claude answers issues here — the butler is
Claude ·
architecture deep-dive ·
One transcript record by uuid, byte-identical — the exact original bytes. Use to recover a specific tool result or turn that a summary or thread view elided.
| Name | Required | Description | Default |
|---|---|---|---|
| uuid | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Given no annotations, the description adds value by disclosing 'byte-identical — the exact original bytes,' indicating read-only, deterministic behavior. It does not mention permissions or error handling, but for a simple retrieval, this is nearly sufficient.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
A single, well-structured sentence with no filler. The core information is front-loaded: identity, purpose, and key characteristic (byte-identical).
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
No output schema is provided, and the description does not explain the return format beyond 'byte-identical bytes'. It omits error scenarios (e.g., missing uuid) and any structural details of the record. The tool is simple, but completeness is lacking.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
With 0% schema coverage, the description must explain the uuid parameter, but only states 'by uuid'. It does not describe what the uuid represents (e.g., transcript record UUID) or its format, leaving the agent to infer.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states it retrieves a single transcript record by uuid and returns the exact original bytes. It specifically differentiates from siblings like fable_thread and fable_recall by positioning itself as the tool for raw, unmodified data.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly tells when to use this tool: to recover a specific tool result or turn that a summary or thread view elided. Implicitly advises against using it when a summary or thread view is sufficient, naming alternatives.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Auto-assemble a paste-ready context pack for a task: searches the archive, picks the strongest threads, splits the budget across them. Returns one sentinel-wrapped block.
| Name | Required | Description | Default |
|---|---|---|---|
| query | Yes | ||
| budget | No | ||
| max_threads | No | ||
| project | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the burden. It discloses the key behavior: searching, selection, budget splitting, and sentinel-wrapped output. However, it does not state whether the tool is read-only or if it modifies any state, and lacks details on side effects.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single sentence that efficiently conveys the tool's purpose, mechanism, and output. Every part is informative with no redundancy or filler.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given no annotations, no output schema, and sparse schema descriptions, the description covers the core idea but lacks specifics (e.g., what 'strongest threads' means, budget units, wrapping format). It is minimally adequate but leaves gaps for an agent to infer.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 0%, so the description must compensate. It adds meaning for 'query', 'budget', and 'max_threads' by describing their role in the process, but does not mention the 'project' parameter, leaving it unexplained.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool auto-assembles a paste-ready context pack for a task, involving searching the archive, picking strongest threads, and splitting budget. It differentiates from siblings like fable_search or fable_recall by focusing on compilation, though it doesn't explicitly contrast with them.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description gives no guidance on when to use this tool versus alternatives. It does not mention prerequisites, exclusions, or compare to sibling tools, leaving the agent to infer usage context.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Unified diff between any two reconstructed versions of a file (version indices from fable_file_history) — recover exactly what changed between two past edits, or between a past version and the latest. Pass the file path and the two version indices a and b.
| Name | Required | Description | Default |
|---|---|---|---|
| path | Yes | ||
| a | Yes | before version index | |
| b | Yes | after version index |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description carries full burden. It mentions 'unified diff' which implies a standard output format, but does not explicitly state that the operation is read-only, nor does it disclose any potential side effects. The description is adequate but could be more explicit about behavior.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is two sentences long, front-loaded with the primary purpose, and contains no superfluous information. Every sentence is essential and clear.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple diff tool with three parameters and no output schema, the description is largely complete. It specifies inputs (path, version indices) and output format (unified diff), and references the companion tool for obtaining indices. It could briefly describe the output format, but 'unified diff' is sufficiently standard.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The schema covers 2 of 3 parameters with descriptions (a and b). The description adds context by linking the version indices to the sibling tool fable_file_history, but does not elaborate on the 'path' parameter beyond what is obvious. This adds moderate value beyond the schema.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description explicitly states the tool computes a 'unified diff between any two reconstructed versions of a file', using a specific verb and resource. It references version indices from a sibling tool (fable_file_history), clearly distinguishing it from other file-related tools like fable_file_history or fable_files.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description gives concrete usage scenarios: 'recover exactly what changed between two past edits, or between a past version and the latest'. It also instructs to pass 'the file path and the two version indices a and b'. While it does not explicitly exclude alternatives, the context makes it clear when to use this tool.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
EVERY version of a file Claude ever edited, reconstructed from the transcript — each version's index, timestamp, tool, session and fidelity (exact replay vs rebuilt-backward). Use to see how a file evolved, or to find the two version indices to diff. Pass a file path (or a distinctive substring of it).
| Name | Required | Description | Default |
|---|---|---|---|
| path | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries full burden. It discloses that versions are 'reconstructed from the transcript' and mentions fidelity types (exact replay vs rebuilt-backward), providing key behavioral traits.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two sentences, front-loaded with key information, no redundancy. Every sentence adds value.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Covers version details and fidelity, but lacks mention of error handling or return format. Given no output schema, more could be said about what is returned when a file is not found.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The only parameter 'path' is enhanced beyond schema (which has no description) by noting it accepts 'a file path (or a distinctive substring of it),' adding crucial matching semantics.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Description clearly states it shows every version of a file Claude edited, reconstructed from transcript, with details like index, timestamp, tool, session, and fidelity. It distinguishes itself from siblings like fable_file_diff by focusing on history rather than diffing.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly states when to use: 'see how a file evolved, or to find the two version indices to diff.' It implies the complementary use with fable_file_diff, providing context for appropriate usage.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
List the files Claude has edited — across the whole archive, or within one session — with edit/write counts and last-touched time. Use to DISCOVER what a past session changed before pulling a file's history. Filter by a path substring or a session id.
| Name | Required | Description | Default |
|---|---|---|---|
| query | No | path substring filter, e.g. serve.py | |
| session_id | No | limit to one session's files | |
| limit | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description carries the full burden. It discloses that the tool lists files with 'edit/write counts and last-touched time,' implying a read-only, non-destructive operation. However, it does not discuss performance, authorization needs, or other behavioral traits beyond the basic functionality.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is concise (three sentences) and front-loaded with the primary action. Every sentence adds value: purpose, return info, and usage guidance. No redundant or verbose language.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given 3 simple parameters and no output schema, the description fully covers the tool's functionality, return values, filtering options, and intended use case. It is complete for an agent to understand when and how to use the tool.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
With 67% schema description coverage (query and session_id described in schema, limit not), the description adds value by explaining that the parameters are for filtering ('Filter by a path substring or a session id'). It reinforces the purpose of the parameters beyond the schema descriptions.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description uses specific verbs ('List'), identifies the resource ('files Claude has edited'), and clarifies scope ('across the whole archive, or within one session'). It also distinguishes from sibling tool fable_file_history by indicating when to use this tool before pulling history.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description explicitly states the primary use case: 'Use to DISCOVER what a past session changed before pulling a file's history.' It also explains filtering by path substring or session id, providing clear context for when to use this tool, though it doesn't explicitly state alternatives or when not to use.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Slim a session transcript NOW (tool noise, images, bloat) with a vault backup sealed first — nothing is lost, everything stays recallable. Use when the user asks to prune/slim a session or complains about context size. After pruning the CURRENT session, tell the user to /exit and run the returned resume command to load the slim version.
| Name | Required | Description | Default |
|---|---|---|---|
| session_id | Yes | session to prune (the current session's id works — the rewrite is atomic and append-safe) | |
| strip_images | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations exist, so description carries full burden. It discloses key behaviors: creates a vault backup before pruning, atomic and append-safe rewrite, and returns a resume command. Could elaborate on vault mechanism but sufficient for agent understanding.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Extremely concise: two sentences covering purpose, usage, and post-instructions. Front-loaded with action and key details. No redundant phrases.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a 2-param no-output tool, the description covers what it does, when to use it, and the backup behavior. Missing details on output format (the resume command) and vault backup mechanics. Good but not exhaustive.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 50% with session_id described well (atomic, append-safe). The description adds context that strip_images removes images, tying the boolean parameter to the stated goal of slimming. Adds value beyond schema by explaining the backup and atomicity.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool prunes a session transcript by removing tool noise, images, and bloat. The verb 'prune' is specific and distinct from sibling tools (e.g., fable_recall, fable_block), making it easy for an agent to select this tool.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly says 'Use when the user asks to prune/slim a session or complains about context size.' It also gives post-usage instructions (tell user to /exit and run resume command). Lacks when-not-to-use or alternatives, but provides clear context.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Read the durable facts the user stored via /remember (the READ side of fable_remember) — lasting preferences, decisions and constraints, the same ones auto-injected at session start. Call to re-check what the user has committed to before assuming. Optionally scope to a project.
| Name | Required | Description | Default |
|---|---|---|---|
| project | No | scope to a project (omit for all/global) |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations provided, so description carries full burden. States it is a read operation (non-destructive), mentions data is durable and auto-injected at session start. Does not cover auth or rate limits, but sufficient for a read-only recall tool.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Four sentences, no wasted words, but the first sentence is somewhat long. Still efficiently conveys purpose and usage.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
No output schema, but description clearly explains what is returned (durable facts, preferences, decisions, constraints) and notes auto-injection. Parameter is fully covered. Complete for a recall tool.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, and description adds the meaning of the optional 'project' parameter: 'scope to a project'. Baseline 3 is adjusted upward for the added context.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Clearly states the verb 'Read' and the resource 'durable facts' stored via /remember, with specific examples (preferences, decisions, constraints). Distinguishes from sibling fable_remember which is the write side.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly says when to use: 'Call to re-check what the user has committed to before assuming.' Implicitly differentiates from fable_remember (write). However, no explicit mention of when not to use or alternatives like fable_context.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Store a durable fact the user wants remembered across all future sessions (auto-injected at session start). Use when the user says 'remember that...' or states a lasting preference/decision.
| Name | Required | Description | Default |
|---|---|---|---|
| fact | Yes | ||
| project | No | scope to a project (omit for global) |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations provided, so description carries full burden. Discloses durability and auto-injection at session start, but does not mention overwrite behavior or limits. Adequate for a simple tool.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two sentences: first states core function, second gives usage guidance. No fluff, front-loaded, every sentence earns its place.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
No output schema, no annotations. Description covers persistence and injection but lacks details on limits (e.g., character cap), overwrite behavior, or return value. Adequate for simple tool but has gaps.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage 50%: 'project' has a description in schema, 'fact' does not. Tool description implies 'fact' is the fact to remember but adds no extra detail. Description provides marginal value over schema.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Description clearly states verb 'store', resource 'durable fact', and scope 'across all future sessions with auto-injection'. Distinguishes itself from siblings by emphasizing persistence and injection behavior.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly says 'Use when the user says "remember that..." or states a lasting preference/decision.' Provides clear usage context but lacks explicit when-not-to-use or alternatives, though siblings like fable_context are implied.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
RECALL EARLIER CONVERSATION. Use this WHENEVER you need context that may be outside your window — after a compaction, deep in a long session, or to recall a past decision / discussion across ANY session. PREFER IT over guessing or trusting a compaction summary (the summary is lossy; this is the exact archive). Returns ranked threads with ids, turn/token counts, card titles, decisions and outcomes; then call fable_thread to read one verbatim.
| Name | Required | Description | Default |
|---|---|---|---|
| query | Yes | ||
| operative | No | action-verb facet, e.g. decide, fix | |
| target | No | file/crate/identifier facet | |
| project | No | ||
| kind | No | ||
| tag | No | taxonomy filter 'family:value' (e.g. 'topic:auth', 'decision:architecture', 'technology:rust'); a bare value matches any family. DISCOVER valid tags first via fable_tags. | |
| sort | No | ||
| limit | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description carries the full burden. It discloses that the tool returns ranked threads with ids, turn/token counts, card titles, decisions, and outcomes. It recommends calling fable_thread to read a thread verbatim, implying a read-only operation. No mention of destructive behavior or rate limits, but the description gives sufficient behavioral context for an agent.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single block but front-loaded with the essential phrase 'RECALL EARLIER CONVERSATION.' It uses capitalization to highlight key concepts. Every sentence adds value, though it could be broken into more structured bullets. Overall, it is efficient and not verbose.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the complexity of 8 parameters, no output schema, and no annotations, the description provides a clear workflow: search and then call fable_thread to read. It explains the return format sufficiently. Some details like pagination or error handling are missing, but for a search tool, the description is fairly complete.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is only 38% (3 of 8 parameters have schema descriptions). However, the tool description adds meaning for several parameters: it explains the 'operative' parameter as an 'action-verb facet,' 'target' as a 'file/crate/identifier facet,' and 'tag' as a 'taxonomy filter.' It also explains the overall workflow. This compensates for the missing schema descriptions to a large extent.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool is for recalling earlier conversations. It distinguishes itself from sibling tools by explaining the workflow: search first, then call fable_thread to read a thread verbatim. The phrase 'RECALL EARLIER CONVERSATION' is a specific verb+resource that sets clear expectations.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly tells when to use the tool: 'whenever you need context that may be outside your window,' after a compaction, in a long session, or to recall past decisions. It also advises to prefer this tool over guessing or using a compaction summary, noting that the summary is lossy. This provides excellent guidance on when and when not to use the tool.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
DISCOVER the taxonomy tags fable assigns to threads, for precise tag-filtered recall. Call with NO args to list the tag FAMILIES (domain, activity, topic, technology, pattern, intent, outcome, decision…) with counts; call with family='' to list THAT family's values. Then pass tag='family:value' to fable_search to scope recall to exactly that kind of work — progressive disclosure, so you never need the whole taxonomy up front.
| Name | Required | Description | Default |
|---|---|---|---|
| family | No | omit to list families; pass one (e.g. 'topic') to list its tag values |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description fully discloses the read-only behavior without annotations: it lists tag families or values, mentions it returns counts, and uses progressive disclosure. No side effects or destructive actions are implied or hidden.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is concise, front-loading the purpose, and uses clear sentences. However, it could be slightly more compact; 'progressive disclosure' and the integration note add necessary context but slightly extend length.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Despite lacking an output schema, the description provides sufficient detail about return values (families with counts, family values) and usage flow. It covers the main use case and integration, though it does not specify the exact data format.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The input schema already covers the parameter well (100% coverage), but the description adds valuable context by explaining the dual mode (list families vs. values) and the progressive disclosure workflow with fable_search. It goes beyond the schema's simple description.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool's purpose: 'DISCOVER the taxonomy tags fable assigns to threads, for precise tag-filtered recall.' It specifies two modes (list families or list values within a family) and distinguishes from sibling tools like fable_search by focusing on taxonomy discovery rather than recall.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicit usage instructions: 'Call with NO args to list the tag FAMILIES...; call with family='<one>' to list THAT family's values.' It also explains how to use the results with fable_search ('pass tag='family:value' to fable_search to scope recall'), providing clear when-to-use and integration guidance.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Read one conversation thread VERBATIM (user → assistant → tool turns, in order) under a token budget — the exact past turns, not a paraphrase. Use after fable_search (or with a known prompt_id) to recover precise detail a summary would have lost. Bulky tool results are elided with block pointers (fetch via fable_block).
| Name | Required | Description | Default |
|---|---|---|---|
| prompt_id | Yes | ||
| budget | No | ||
| raw | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Without annotations, the description effectively communicates the tool's read-only behavior: it reads verbatim ordered turns under a token budget, and truncates bulky results with block pointers. This adequately discloses key behavioral traits beyond what annotations would provide.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is two sentences, front-loading the core function and following with essential usage guidance. Every sentence adds value without redundancy or unnecessary information.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given zero schema descriptions, no output schema, and no annotations, the description covers the main purpose and usage context but lacks details on the 'raw' parameter, output format, and exact budget truncation behavior. It is adequate but not fully complete for an agent to predict all behaviors.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The input schema has 3 parameters with 0% description coverage, so the description must compensate. It mentions 'token budget' (budget) and hints at truncation, but does not explain the 'raw' parameter or provide details on prompt_id usage beyond being required. The description adds minimal semantic value beyond the schema.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the verb 'read' and the resource 'conversation thread ver batim', specifying it returns exact past turns, not a paraphrase. It distinguishes itself from fable_search by indicating that fable_search returns summaries, making the tool's unique purpose explicit.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description explicitly advises using this tool after fable_search or with a known prompt_id to recover detail lost in summaries. It also mentions that bulky results are elided with block pointers (fetch via fable_block). This provides clear context on when and how to use the tool, though it could be more explicit about when not to use it.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections. Dates show when Glama detected each change.
v0.1.0fable_blockfable_contextfable_file_difffable_file_historyfable_filesfable_prunefable_recallfable_rememberfable_searchfable_tagsfable_threadEach tool targets a distinct operation: block retrieval, context assembly, file diff/history, file listing, session pruning, fact storage/recall, search, tag discovery, and thread reading. There is no overlap between these functions.
All tools follow the 'fable_<function>' pattern with snake_case. The naming is uniform and predictable, making it easy for an agent to infer the tool's purpose from its name.
With 11 tools, the server covers a complex domain—conversation and file history recall—without being overwhelming. Each tool serves a clear role, and the count feels justified.
The tool set covers the main workflows: storing/recalling facts, searching threads, managing file history, and pruning sessions. Minor gaps include the lack of a tool to update or delete remembered facts, but the core functionality is well-covered.
Unclaimed servers have limited discoverability.
If you are the server author, to access and configure the admin panel.
Persistent context for Claude. Your AI always knows your projects and next actions across sessions.
Persistent memory for Claude Code and Cursor. Stop re-explaining your project every session.
Persistent memory layer for AI tools. Save and recall notes across Claude and other MCP clients.
Persistent AI memory shared across Claude, ChatGPT, coding agents, and compatible MCP clients.
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/grooverLab/fable'
If you have feedback or need assistance with the MCP directory API, please join our Discord server
