Provides a tool to delegate prompts to OpenAI's Codex CLI, enabling coding tasks to be executed by Codex, and can use Codex as a runner or reviewer in change reviews.
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., "@agent-mcp-hub Run all agents in parallel: explain the retry logic in src/exec.ts"
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.
One MCP server that bridges multiple CLI coding agents — Codex, Cursor, OpenCode, Claude, and Antigravity — into any MCP client.
stdio only — by design. The hub ships no Docker image and no HTTP transport (both were removed during the 0.5.x line). A containerised or remote server cannot see the caller's repository path and cannot reuse the caller's CLI logins — it would break the product contract on both halves. The hub runs as a child process of your MCP client, on your machine, as you.
Tool | Description |
| Delegate a prompt to |
| Delegate a prompt to |
| Delegate a prompt to |
| Delegate a prompt to the Claude Code CLI (prompt piped via stdin) |
| Delegate a prompt to Google Antigravity ( |
| Same prompt to all agents in parallel, results side by side |
| Which agent CLIs are installed and on PATH |
| Health check |
Agent tools accept prompt (required), model, cwd, timeoutMs (total runtime
cap, default 1800000 = 30 min), and idleTimeoutMs (inactivity cap, default 300000
= 5 min). See Long-running tasks & timeouts.
Known limitation: opencode prompts may not start with - (its CLI could parse
them as flags); the tool returns an actionable error instead of guessing.
Model ids are agent-specific and do not overlap. o3 means nothing to
agy, gemini-3.6-flash-low means nothing to codex, and opencode namespaces
its own (opencode/big-pickle). So a single model value is valid for at most
one agent in a fan-out.
run_all takes models — a per-agent map. Agents you don't list fall back
to model, then to their own CLI default.
review_change takes runnerModel and reviewerModel, so the agent
writing the change and the agent judging it can use different engines and
tiers. model remains a fallback for both.
{
"tool": "run_all",
"arguments": {
"prompt": "Explain the retry logic in src/exec.ts",
"models": { "codex": "o3", "agy": "gemini-3.6-flash-low", "claude": "haiku" }
}
}An unknown agent name in models is rejected with the list of enabled agents —
a typo never silently falls back to the default model.
Where to find valid ids: opencode models, cursor-agent models and
agy models list them. codex and claude have no such command — codex reads
its default from ~/.codex/config.toml, and claude accepts the documented
aliases (opus, sonnet, haiku, fable). Note claude models is not a
subcommand: it is treated as a prompt, so its "model list" is generated text,
not a source of truth.
When a wrapped CLI fails, the hub classifies the failure and returns a clean,
ANSI-free, actionable isError result — never a raw terminal dump — naming the
class and the exact fix:
Class | Example remediation |
| install the CLI (e.g. |
|
|
| set a model/provider in the CLI's config |
| raise |
| the agent reached the network but its stream keeps dropping (e.g. |
| retry shortly (upstream rate-limit, or the local agent-spawn queue is full) |
| generic non-zero exit — the message includes |
For example, an unauthenticated cursor no longer returns its ANSI "press any
key to sign in" banner — it returns cursor is not authenticated … Fix: run cursor-agent login``.
review_change)Runs a runner agent in a git cwd to make a change, captures the actual
git diff of what changed, then has a reviewer agent judge that diff. Returns
the runner's output, the diff (--stat), and a PASS / WARN / FAIL verdict
with findings.
Inputs: runner, reviewer (agent names), prompt, cwd (must be a git
worktree), optional runnerModel, reviewerModel, model (fallback for both),
timeoutMs.
Key notes:
Cross-agent by design — e.g. codex writes, claude reviews.
Returns the concrete diff that the plain agent tools don't expose.
Newly-created (untracked) files are surfaced to the reviewer with their
contents (bounded: 64 KiB per file, 50 files; excess is truncated and
flagged). git diff alone would omit them entirely.
If the worktree was already dirty, the diff may include pre-existing changes (noted in the output).
Complements — does not replace — client-side stop-hooks or PR-time CI review.
The confirm gate (MCP_CONFIRM) applies.
The reviewer runs least-privilege — but how much that guarantees depends on the agent. The reviewer's prompt embeds an attacker-influenced diff, so it is run read-only rather than with the write grant the runner gets. Only two of the five are real restrictions:
Reviewer | Read-only mechanism | Enforced? |
|
| Yes — OS-level sandbox |
|
| Yes — harness-level deny |
|
| No — model-advisory only |
| permission grant withheld | Partly — headless auto-deny, not a sandbox |
| none available | No — unrestricted |
Pick codex or claude as the reviewer if you want the restriction to be real.
This is defense-in-depth alongside the nonce-fenced diff and the throwaway temp
cwd; none of it is a sandbox.
{
"tool": "review_change",
"arguments": {
"runner": "codex",
"reviewer": "claude",
"prompt": "Add retry with exponential backoff to the API client",
"cwd": "/Users/you/projects/my-app"
}
}Related MCP server: personal-mcp
Install and authenticate the CLIs you want to use (any subset works):
Codex: npm i -g @openai/codex && codex login
Cursor: curl https://cursor.com/install -fsS | bash && cursor-agent login
OpenCode: npm i -g opencode-ai && opencode auth login
Claude Code: npm i -g @anthropic-ai/claude-code && claude (first run logs in)
Antigravity: install the Antigravity CLI (e.g. brew install --cask antigravity-cli),
then run agy once and sign in — it has no login subcommand and stores
credentials in the OS keyring, so there is no API-key env var to set
Recommended — global install (fast, reliable startup): install the pinned
version once, then point your client at the agent-mcp-hub binary. Startup is
instant and the client connects reliably.
npm i -g agent-mcp-hub@0.5.0claude mcp add agent-hub -- agent-mcp-hub{
"mcpServers": {
"agent-hub": {
"command": "agent-mcp-hub"
}
}
}No global install, but npx re-resolves the package on every launch, so first start is slower and can occasionally trip a client's connection-probe timeout (the server itself is fine — just retry). Prefer the global install for a persistent setup.
claude mcp add agent-hub -- npx -y agent-mcp-hub@0.5.0
# mcp.json: "command": "npx", "args": ["-y", "agent-mcp-hub@0.5.0"]Pre-release / fallback: To test an unreleased commit, run directly from GitHub:
npx -y github:blackaxgit/agent-mcp-hub#<tag-or-sha>. This builds from source on first fetch, so under npm v12+ you must allow thepreparescript.
MCP_AGENTS — comma-separated allowlist of the agents to expose
(codex,cursor,opencode,claude,agy). Unset or empty exposes all agents. Disabled
agents get no tool and are absent from list_agents/run_all. An unknown name
fails at startup with an error listing the valid names, so typos never silently
disable an agent.
For stdio, set it in the client's mcp.json:
{
"mcpServers": {
"agent-hub": {
"command": "npx",
"args": ["-y", "agent-mcp-hub@0.5.0"],
"env": { "MCP_AGENTS": "codex,claude" }
}
}
}MCP_CONFIRMSet MCP_CONFIRM=1 (values 1/true/on/all; default off) to require a
confirmation before any agent tool — and run_all — actually spawns a CLI. The
server sends a brief summary (agent · prompt · cwd · model) and waits: accept
runs the agent, decline runs nothing and returns a terminal cancellation.
This uses the standard MCP elicitation capability, so it is client/IDE-agnostic — it works with any MCP client that supports form elicitation (Claude Code, Cursor, VS Code, Zed, Windsurf, custom SDK clients, …); the gate keys on the protocol capability, never a product name. Clients that don't support elicitation transparently run without a prompt (no hang, no error).
{
"mcpServers": {
"agent-hub": {
"command": "npx",
"args": ["-y", "agent-mcp-hub@0.5.0"],
"env": { "MCP_CONFIRM": "1" }
}
}
}A complex agent task can run for many minutes. The hub bounds each run with two independent timers so a productive long run survives while a genuinely stuck one fails fast:
Idle (inactivity) timeout — idleTimeoutMs (per call) / MCP_AGENT_IDLE_TIMEOUT_MS
(env), default 300000 (5 min). The timer resets on every chunk of output the
CLI produces, so an agent that keeps working (streaming output) never trips it. An
agent that goes silent — e.g. opencode stuck on an unreachable model backend —
is killed after the idle window with an actionable "no output — the agent may be
hung or its model/backend is unreachable" error, instead of burning the full cap.
Total runtime cap — timeoutMs (per call) / MCP_AGENT_TIMEOUT_MS (env),
default 1800000 (30 min). A hard upper bound regardless of activity.
Whichever fires first kills the agent's process group. Tradeoff: the idle reset
assumes the CLI streams intermediate output. codex and opencode do; claude -p,
cursor-agent -p and agy --print may emit only the final result, so a long
silent task on those can be idle-killed at 5 min — raise idleTimeoutMs /
MCP_AGENT_IDLE_TIMEOUT_MS for such tasks, or rely on the total cap.
agy is the strongest case here: it emits its entire answer in one flush at the
end (measured — every line of a 12-line reply arrived at the same instant), and
the hub raises its internal --print-timeout to 48h so that agy's own 5-minute
default can never truncate a run. The idle timer is therefore agy's only
protection against a genuine hang; budget idleTimeoutMs accordingly for long
autonomous agy tasks.
While an agent runs, the hub emits MCP progress notifications to clients that
request them (_meta.progressToken) — live feedback during long runs. Note: on
Claude Code (stdio) the per-server request timeout in .mcp.json (or the
MCP_TOOL_TIMEOUT env var it honors) is a hard wall-clock that progress does
not reset (default ~28h) — raise it if you lowered it below your longest run.
Global install: install the new version — the pinned agent-mcp-hub command
in your MCP config picks it up on the next client start:
npm i -g agent-mcp-hub@0.6.0npx (pinned): bump the pinned version in your MCP config — e.g. change
agent-mcp-hub@0.5.0 to agent-mcp-hub@0.6.0 everywhere.
Always-latest (not recommended for shared configs): use agent-mcp-hub@latest
instead of a pinned version. Note that npx caches by version — it may serve a
stale copy. Force a fresh fetch with npx --prefer-online agent-mcp-hub or
npx clear-npx-cache.
Pinning is reproducible and recommended for team-wide or checked-in
mcp.json files.
npm install
npm test # vitest
npm run typecheck # strict TS over src + tests
npm run dev # run from source over stdio
npm run build # emit dist/Pure adapters (src/adapters/* — prompt → {args, stdin?}, no I/O) → one
subprocess boundary (src/exec.ts) → MCP stdio server (src/server.ts). Adding
an agent = one ~15-line adapter file + one line in src/registry.ts.
Mozilla Public License 2.0 (MPL-2.0).
Versions before 0.5.2 were released under the MIT license; that grant stands for those versions. MPL-2.0 applies from 0.5.2 onward.
Claude Code — Anthropic's coding agent for complex implementation, refactoring, and code review. Runs the claude CLI non-interactively in cwd — it can read and edit files there and may take time or use the agent's own model quota — and returns its output. On common failures returns a classified, actionable error (not installed / not authenticated with the exact login command / not configured / timed out / busy / output-limit); other non-zero exits return a clipped stderr/stdout tail. Check availability with list_agents first.
| Name | Required | Description | Default |
|---|---|---|---|
| cwd | No | Working directory for the CLI. Prefer an absolute path; a relative path resolves from the server process's cwd. Not a sandbox — the agent may read/edit any files it can access. | |
| model | No | Optional model id passed through to the CLI, overriding that CLI's configured/default model (e.g. "o3"). Model names are agent-specific. Must match [A-Za-z0-9][A-Za-z0-9._:/-]{0,127} — a flag-shaped value is rejected. | |
| prompt | Yes | The task or question for the agent, in natural language. | |
| timeoutMs | No | Total runtime cap in milliseconds — the hard upper bound on the whole run once the CLI starts (excludes time queued behind the concurrency limit); the process group is killed if exceeded (default 1800000 = 30 minutes). | |
| idleTimeoutMs | No | Idle/inactivity timeout in milliseconds — the run is killed only if the agent produces NO output for this long (the timer resets on each output chunk; default 300000 = 5 minutes). |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Beyond the annotations (readOnlyHint=false, openWorldHint=true, destructiveHint=true), the description discloses the non-interactive execution, file read/edit capabilities, potential time and quota usage, and a detailed error-handling scheme (classified errors for common failures, clipped tail for others). This significantly enriches the agent's understanding of side effects and limitations.
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 four sentences, each carrying distinct information: what the tool is, how it executes, how failures are handled, and a prerequisite. There is no redundancy, and the most important facts are front-loaded.
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?
The description covers purpose, execution, error behavior, and prerequisites, which is quite complete for a complex tool with no output schema. It briefly explains the return value ('returns its output' and failure modes), though a more detailed description of the successful output format would make it fully 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 100%, so the baseline is 3. The description mentions 'cwd' contextually but does not add semantics beyond what the schema already provides; the schema itself fully describes parameters like timeoutMs, model, and prompt.
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 function: 'Runs the `claude` CLI non-interactively in `cwd`' for 'complex implementation, refactoring, and code review'. This specifies both the verb and resource, and identifies it as 'Anthropic's coding agent', distinguishing it from sibling tools like codex or run_all.
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?
It provides clear context for when to use: for complex coding tasks requiring implementation, refactoring, or code review. It also gives a specific prerequisite ('Check availability with list_agents first') and warns about time and model quota costs, which helps the agent decide. However, it does not explicitly contrast with alternatives like codex, so it lacks explicit exclusions.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
OpenAI Codex — a terminal coding agent (GPT/o-series) that autonomously reads and edits files to implement, refactor, and fix code. Runs the codex CLI non-interactively in cwd — it can read and edit files there and may take time or use the agent's own model quota — and returns its output. On common failures returns a classified, actionable error (not installed / not authenticated with the exact login command / not configured / timed out / busy / output-limit); other non-zero exits return a clipped stderr/stdout tail. Check availability with list_agents first.
| Name | Required | Description | Default |
|---|---|---|---|
| cwd | No | Working directory for the CLI. Prefer an absolute path; a relative path resolves from the server process's cwd. Not a sandbox — the agent may read/edit any files it can access. | |
| model | No | Optional model id passed through to the CLI, overriding that CLI's configured/default model (e.g. "o3"). Model names are agent-specific. Must match [A-Za-z0-9][A-Za-z0-9._:/-]{0,127} — a flag-shaped value is rejected. | |
| prompt | Yes | The task or question for the agent, in natural language. | |
| timeoutMs | No | Total runtime cap in milliseconds — the hard upper bound on the whole run once the CLI starts (excludes time queued behind the concurrency limit); the process group is killed if exceeded (default 1800000 = 30 minutes). | |
| idleTimeoutMs | No | Idle/inactivity timeout in milliseconds — the run is killed only if the agent produces NO output for this long (the timer resets on each output chunk; default 300000 = 5 minutes). |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already flag destructive and open-world behavior, but the description adds rich context: non-interactive execution, ability to read/edit any accessible files (not sandboxed), time/model quota usage, classified failure modes, and clipped stderr/stdout on non-zero exits. No contradiction with annotations.
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?
Three dense sentences: purpose, execution behavior, failure handling, and availability check. Information is front-loaded with tool identity, and every sentence earns its place without fluff.
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 no output schema, the description covers success output ('returns its output'), failure classification, and operational caveats. Combined with strong schema and annotations, the tool is fully specified for an agent to use safely.
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?
All 5 parameters have detailed schema descriptions (100% coverage) covering cwd resolution, model override validation, and timeout semantics. The description itself adds no extra parameter detail, so the baseline 3 applies.
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 opens with a specific verb+resource: 'Runs the codex CLI non-interactively in cwd' and explains it 'autonomously reads and edits files to implement, refactor, and fix code.' This clearly distinguishes it from siblings like claude and run_all by naming the specific agent and its function.
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?
It explicitly instructs to 'Check availability with list_agents first' and clearly implies when to use: when you need OpenAI Codex to autonomously work on code. It does not explicitly compare to claude or run_all, but the use case is clear enough.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
List the wrapped coding-agent CLIs and whether each can actually run (edits nothing). Each entry reports installed (binary resolves on PATH with the exec bit) and usable (a probe succeeded), plus a reason when it cannot run; available mirrors usable. A CLI can be installed but unusable — codex exits 0 from --version even when its home is unwritable and no real run can succeed — so prefer usable. Read-only; call this first to choose an agent before delegating.
| Name | Required | Description | Default |
|---|---|---|---|
No parameters | |||
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Despite the readOnlyHint annotation, the description adds valuable nuance: it explains the exact meaning of installed vs usable, and the concrete pitfall where codex --version exits 0 despite an unwritable home. This goes beyond the annotation's bare safety flag and clarifies what 'can actually run' means.
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?
Every sentence serves a purpose: purpose, field meanings, a critical caveat, and usage direction. It is front-loaded with a clear statement of what the tool does and remains compact despite covering nuanced semantics.
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?
With no output schema, the description fully specifies return fields (installed, usable, reason, available) and their semantics, including the installed-but-unusable distinction. The practical guidance to prefer usable completes the picture for choosing an agent.
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 tool accepts zero parameters, so the schema already fully documents the input. The description focuses instead on return fields, which is appropriate; no parameter clarification is needed beyond the empty 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 opens with a specific verb-resource pair ('List the wrapped coding-agent CLIs') and adds scope ('whether each can actually run'), clearly distinguishing it from sibling agent-execution tools like codex/claude. The elaboration about installed vs usable further pins down its function.
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?
It explicitly instructs 'call this first to choose an agent before delegating,' establishing when to invoke it relative to sibling tools. It also advises 'prefer usable' over installed, giving evaluation guidance for interpretation.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Liveness check for agent-mcp-hub — returns "pong". Read-only, no side effects.
| Name | Required | Description | Default |
|---|---|---|---|
No parameters | |||
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint=true and openWorldHint=false. The description adds the concrete response ('returns "pong"') and reinforces the side-effect-free nature. It doesn't repeat or contradict annotations, and provides sufficient behavioral context for a zero-parameter operation.
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, front-loaded sentence. It states the purpose, the return value, and the safety profile without any redundant or extraneous text.
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 trivial ping tool with no parameters, no output schema, and annotations covering read-only behavior, the description is complete. An agent has everything needed to select and invoke it correctly.
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 tool has zero parameters, so the schema is trivially complete (100% coverage). The description adds no parameter details because none are needed. Baseline 4 is appropriate for a no-parameter tool.
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 a specific verb and resource: 'Liveness check for agent-mcp-hub' and what it returns ('pong'). This distinguishes it from sibling tools like list_agents or run_all, which serve different purposes.
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 context is clear: it's a liveness check, so an agent knows to use it to verify the server is reachable. It doesn't explicitly mention when not to use it or name alternatives, but given the tool's simplicity and distinct purpose among siblings, the usage is well implied.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Run the runner agent in cwd (which edits files), capture the concrete git diff of the change, then have the reviewer agent judge it and return a PASS/WARN/FAIL verdict along with the runner output, the diff, and the review. Requires a git worktree. Newly-created (untracked) files are reviewed by their contents too (bounded per file), not just by name. The diff may include pre-existing changes if the worktree was already dirty.
| Name | Required | Description | Default |
|---|---|---|---|
| cwd | Yes | Git working tree the runner operates in (REQUIRED) | |
| model | No | Fallback model for BOTH agents. Only useful when runner and reviewer are the same agent — model namespaces do not overlap, so one value cannot be valid for two different CLIs. For the cross-agent setup this tool recommends, use `runnerModel` / `reviewerModel`. Must match [A-Za-z0-9][A-Za-z0-9._:/-]{0,127}. | |
| prompt | Yes | Task or question to send to the runner agent | |
| runner | Yes | Adapter name of the agent that edits files | |
| reviewer | Yes | Adapter name of the agent that judges the change | |
| timeoutMs | No | Per-agent timeout in ms | |
| runnerModel | No | Model for the RUNNER agent, overriding `model`. Use this (not `model`) when runner and reviewer are different agents. | |
| reviewerModel | No | Model for the REVIEWER agent, overriding `model`. Lets you review with a different engine/tier than the one that wrote the change. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description goes well beyond the annotations by detailing the exact sequence of operations (runner edits → diff captured → reviewer judges → verdict returned). It also discloses specific edge-case behaviors: untracked files are reviewed by contents (bounded per file) and the diff may include pre-existing dirty changes. This adds substantial value beyond the readOnlyHint/destructiveHint flags.
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 compact and well-structured: three sentences that explain the core workflow, a prerequisite, and two important edge-case caveats. It is front-loaded with the primary action and avoids redundancy, earning its place with no fluff.
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 tool's complexity (multi-agent, 8 params, no output schema), the description covers the key aspects: operational flow, return contents (verdict, runner output, diff, review), and environmental requirements. However, it could more explicitly state that this is a persistent-mutation operation and when to avoid it, though annotations handle the safety flag partially.
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?
All 8 parameters have descriptions in the schema, so the baseline is 3. The tool description reinforces the role of `runner` and `reviewer` (one edits, one judges) but does not add meaning beyond what the schema already states. No additional parameter context is provided 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 specifies the tool's purpose: run the `runner` agent to edit files, capture the `git diff`, then have the `reviewer` agent judge it and return a verdict with outputs. This distinguishes it from sibling tools like `codex` or `run_all` by describing its unique multi-step review workflow.
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 provides clear context for when to use the tool—when you need an agent to make a change and have another agent review it. It also gives a necessary prerequisite ('Requires a git worktree') and warns about pre-existing changes in a dirty worktree, but it does not explicitly name alternatives or spell out 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.
Fan the SAME prompt out to every enabled agent concurrently and return each agent's answer, labelled per agent — for comparing agents or cross-checking a result. Spawns every CLI (each can read/edit files in cwd) so it can be slow or use several agents' quotas; one confirmation covers the whole batch.
| Name | Required | Description | Default |
|---|---|---|---|
| cwd | No | Working directory for every CLI. Prefer an absolute path; a relative path resolves from the server process's cwd. Not a sandbox. | |
| model | No | Optional model id applied to EVERY agent as a fallback. Model namespaces do not overlap between agents, so a single value is usually valid for only one of them — prefer `models` for a fan-out. Must match [A-Za-z0-9][A-Za-z0-9._:/-]{0,127} — a flag-shaped value is rejected. | |
| models | No | Per-agent model overrides, e.g. {"codex":"o3","agy":"gemini-3.6-flash-low"}. Takes precedence over `model` for the agents named; any agent not listed falls back to `model`, then to its own CLI default. Keys must be enabled agent names (an unknown name is rejected with the valid list). Each value must match [A-Za-z0-9][A-Za-z0-9._:/-]{0,127}. | |
| prompt | Yes | The task or question to send to every agent, in natural language. | |
| timeoutMs | No | Per-agent TOTAL runtime cap in milliseconds once each CLI starts (process group killed if exceeded; default 1800000 = 30 minutes). | |
| idleTimeoutMs | No | Per-agent idle/inactivity timeout in milliseconds — killed only if the agent produces NO output for this long (resets on each output chunk; default 300000 = 5 minutes). |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already indicate destructive/open-world behavior, and the description supplements this with concrete details: 'each can read/edit files in cwd' and 'one confirmation covers the whole batch.' It also mentions performance and quota impacts. This adds value beyond the provided annotations without contradiction.
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 the primary action, followed by a purpose clause and a caveat-rich second sentence. Every phrase earns its place with no redundancy or fluff.
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?
The description covers the core purpose, return format, file access, confirmation behavior, and performance implications. Given the rich schema and annotations, this is a solid overview. It does not mention every detail (e.g., model fallback rules or timeout mechanics), but those are fully documented in the schema, so the description is sufficiently 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 coverage is 100%, and each parameter already has a detailed description (e.g., cwd is 'not a sandbox', model formatting rules, timeout semantics). The tool description only references cwd behaviorally but doesn't add new meaning beyond the schema. Baseline 3 is appropriate.
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 function: 'Fan the SAME prompt out to every enabled agent concurrently and return each agent's answer, labelled per agent.' It uses a specific verb ('fan out') and resource ('every enabled agent'), and distinguishes this from sibling tools like codex/claude, which target a single agent.
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 context on when to use the tool: 'for comparing agents or cross-checking a result.' It also warns about costs: 'can be slow or use several agents' quotas.' However, it does not explicitly name an alternative tool for single-agent use, so it falls just short of a perfect 5.
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.
v0.6.1claudecodexlist_agentspingreview_changerun_allEach tool has a clearly distinct purpose: ping for liveness, list_agents for availability, codex and claude for running specific CLIs, run_all for fan-out, and review_change for the review workflow. Even though codex and claude are similar, the tool name explicitly selects the target CLI, eliminating ambiguity.
Naming is a mix: verb_noun (list_agents, review_change), product names (codex, claude), a single word (ping), and run_all. While readable and following a lowercase snake_case pattern, the lack of a consistent verb_noun style and the use of proper nouns for some tools makes the naming somewhat inconsistent.
With 6 tools, the server is well-scoped for its purpose of wrapping coding-agent CLIs. Each tool addresses a distinct need—health check, discovery, individual execution, fan-out, and review—without redundancy or bloat.
Core workflows are covered: running codex or claude, running all agents, and reviewing changes. However, list_agents implies the server can discover arbitrary agents, yet there is no generic 'run agent by name' tool, only the hardcoded codex and claude. This gap could prevent users from running other agents that list_agents might report as available.
Unclaimed servers have limited discoverability.
If you are the server author, to access and configure the admin panel.
Real-time chat hub for AI agents — Claude Code, Cursor, Cline, Codex over MCP or REST.
Real-time chat for AI agents. Claude Code, Cursor, Cline and Codex join channels over MCP.
One MCP endpoint for Claude, GPT & Gemini: 100+ tools + no-code connectors + agent workers.
Talk to your public-facing AI from any MCP client — Claude, ChatGPT, Cursor, Cline, Windsurf.
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/blackaxgit/agent-mcp-hub'
If you have feedback or need assistance with the MCP directory API, please join our Discord server
