This server connects AI assistants to Microsoft's official MSRC Security Update Guide, enabling natural-language queries over Patch Tuesday releases. All data comes from public sources (MSRC CVRF API, FIRST.org EPSS, CISA KEV) with no API keys required.
Core capabilities:
Monthly overview – Get a summary of any month's security updates, including stats by severity, impact, product family, and exploitation status.
CVE lookup – Full details for a specific CVE: affected products, KB articles, CVSS, EPSS score, KEV status, FAQs, and reference links.
KB lookup (single or batch) – Map up to 30 KB articles to the CVEs they fix, including supersedence chains and known issues before deployment.
Exploitation & threat intelligence – Filter by exploited-in-the-wild, CISA KEV (with federal due dates), EPSS probability, ransomware campaigns, or Microsoft's exploitation likelihood.
Product & severity filtering – Narrow by product name, product family, custom watchlists/profiles (identity-core, endpoint, server-infrastructure), severity, or CVSS score.
CVSS exposure filtering – Filter by attack vector, privileges required, user interaction, and scope to isolate zero-click, network-reachable criticals.
Zero-day & weakness tracking – Find publicly disclosed vulnerabilities or filter by CWE ID/name (e.g., CWE-416, "use after free").
Mitigations & workarounds – Surface Microsoft-provided mitigations, workarounds, and will-not-fix guidance for CVEs or KBs.
Historical trend analysis – Search across up to 12 months with per-month trend counts.
Triage reports – Export prioritized results as Markdown briefings or CSV spreadsheets; all results are ranked most-urgent-first (KEV/exploited → EPSS → severity → CVSS).
Deployment planning – Check restart requirements, fixed builds, and supersedence chains.
Release catalog – List all available Patch Tuesday months with release dates.
Cache control – Force-refresh to bypass caches and inspect cache age/TTL for MSRC, EPSS, or CISA KEV data.
Guided triage prompt – A built-in monthly_triage prompt walks through a full analyst triage workflow, optionally scoped to a product profile and month.
Allows GitHub Copilot to search and filter Microsoft security updates, providing information on CVEs, KB articles, severity ratings, and exploited-in-the-wild status.
mcp-name: io.github.jonnybottles/patch-tuesday
Ask your AI assistant about Microsoft security updates. This Python-based MCP (Model Context Protocol) server connects AI assistants like Claude, Copilot, and ChatGPT to the MSRC Security Update Guide — the authoritative source for every CVE Microsoft patches — enabling natural-language queries over Patch Tuesday releases: CVEs, KB articles, severity ratings, CVSS scores, affected products, and exploited-in-the-wild status.
Patch Tuesday MCP Server bridges Microsoft's official CVRF security update API and your AI assistant, allowing you to:
Get the monthly rollup - "What did this month's Patch Tuesday fix?"
Find what's actively exploited - "Which vulnerabilities are being exploited in the wild?"
Look up any CVE - "Tell me about CVE-2026-41108" (KBs, affected products, CVSS, description)
Map KBs to CVEs - "Which vulnerabilities does KB5094123 fix?"
Filter by product - "What Critical CVEs affect Windows Server 2022 this month?"
Track zero-days - "Were any publicly disclosed vulnerabilities patched in April?"
See what's confirmed exploited - "Which of this month's CVEs are on the CISA KEV list?" — with federal remediation due dates
Rank by exploitation probability - "Show me CVEs with EPSS above 50%" — daily FIRST.org exploit prediction scores
Find zero-click, internet-reachable criticals - "Which Critical CVEs are network-reachable with no privileges and no user interaction?" — filter on the parsed CVSS attack vector, privileges, and user-interaction fields
Jump straight to authoritative sources - every CVE detail carries ready-to-open MSRC, NVD, EPSS, and (when listed) CISA KEV reference links
Avoid stale patches - "Is KB5087538 superseded by anything newer?" — walks Microsoft's supersedence links
Get mitigations when there's no patch yet - "Are there mitigations or workarounds for CVE-2026-47291?" — surfaces Microsoft's mitigation, workaround, and will-not-fix guidance
Spot trends over time - "How many HTTP.sys CVEs shipped over the last 6 months?" (months_back=6, or start_month/end_month) — aggregates matches across released months with per-month counts
Filter by Microsoft's own exploitation forecast - "Which of this month's CVEs does Microsoft rate 'Exploitation More Likely'?" (exploitation_likely=True)
Spot ransomware-weaponized CVEs - "Which CVEs this month are used in known ransomware campaigns?" (ransomware=True; add include_kev_details=True for the full CISA entry with required actions)
Plan the deployment, not just the priority - "Does KB5094123 require a restart, and what build fixes it?" (include_kb_details=True adds per-KB URLs, fixed builds, supersedence, and restart requirements)
Know what Microsoft says an update breaks - "Any known issues with KB5094126 before I roll it out?" (include_known_issues=True adds each KB's Microsoft-confirmed known issues — symptoms, workarounds, and the resolving update — best-effort from its public support page; published mainly for Windows updates, and the response says explicitly when Microsoft publishes none)
Know what an update changes, not just what it fixes - "What's actually in KB5099536 besides the security fixes?" (include_update_summary=True adds the KB support page's summary and improvements highlights — the change-description content the MSRC security data doesn't carry — from the same page fetch as the known-issues lookup)
Slice a month by weakness class - "Show me this month's use-after-free CVEs" (cwe="CWE-416" or cwe="use after free")
Discover the release catalog - "Which monthly releases are available, and when were they last revised?" (list_months=True)
Export a triage briefing - "Give me this month's Critical CVEs as a Markdown report" or "…as CSV" — a prioritized executive summary and table, or a spreadsheet-ready export (format="markdown" / format="csv")
Force-refresh & check data freshness - "Re-pull this month's data fresh" (force_refresh=True) bypasses the in-process caches; include_freshness=True reports the cache age/TTL of the MSRC document and EPSS/KEV enrichment
Prioritize patching - Results are sorted most-urgent-first: KEV/exploited, then EPSS, then severity, then CVSS
Perfect for security analysts, sysadmins, and IT professionals who triage Microsoft security updates every month — without clicking through the Security Update Guide portal.
Data comes from the official, public MSRC CVRF v3 API. No authentication or API key required.
Related MCP server: CVE Intelligence MCP Server
This is the only MCP server that models the Patch Tuesday release itself. Plenty of MCP servers can look up a CVE — general-purpose vulnerability aggregators fan a known CVE ID out across NVD, OSV, and threat-intel feeds. They answer "tell me about CVE-X". But they have no concept of a monthly Microsoft release, a KB article, or a product family — so they structurally cannot answer the questions a Microsoft shop actually asks on the second Tuesday of every month:
The question you actually have | Generic CVE lookup servers | patch-tuesday-mcp |
"Summarize this month's Patch Tuesday" | ❌ no concept of a release | ✅ rollup + stats in one call |
"What Critical CVEs affect Windows Server 2022 this month?" | ❌ can't filter by Microsoft product | ✅ product & family filtering |
"Which vulnerabilities does KB5094123 fix?" | ❌ no KB awareness | ✅ KB ↔ CVE mapping |
"What's being exploited in the wild right now?" | ⚠️ per-CVE only, if you already know the CVE | ✅ filter the whole month |
"What do I patch first?" | ❌ | ✅ urgency-sorted: exploited/KEV → EPSS → severity → CVSS |
"Which criticals are zero-click and internet-reachable?" | ⚠️ per-CVE CVSS only | ✅ filter the month by parsed CVSS attack vector / privileges / user interaction |
"Tell me about CVE-X" | ✅ (often with more ecosystem data) | ✅ MSRC detail: KBs, builds, supersedence, parsed CVSS, MSRC/NVD/EPSS/KEV links |
Under the hood, the difference is the data source: this server parses the full MSRC CVRF monthly documents — the ProductTree, per-product severity threats, exploitability assessments, and KB remediation chains that per-CVE APIs never expose. That's what makes release-centric questions possible.
Other things it deliberately gets right:
Zero API keys, zero accounts — the MSRC API is public; setup is one uvx command
One tool, not thirty — a single consolidated msrc_search keeps your AI client's context lean and tool selection reliable
Built for the monthly workflow — triage a release, brief your team, prioritize patching, then get on with your life
A free remote instance is available at:
https://patch-tuesday-mcp.agreeabledesert-d0b8e491.eastus2.azurecontainerapps.io/mcpNo account or API key needed. The endpoint serves the same public data as a local install — for heavy use or guaranteed availability, run it locally (below) or self-host your own. Only minimal, anonymized usage data is recorded — see Telemetry & Privacy.
Python 3.11+
An MCP-compatible client (Claude Desktop, Cursor, Claude Code, GitHub Copilot CLI, etc.)
uvx (Recommended)If you are installing or running the server via uvx, you must have uv installed first.
uv (includes uvx): https://github.com/astral-sh/uv
Install uv:
# macOS / Linux
curl -LsSf https://astral.sh/uv/install.sh | sh
# Windows (PowerShell)
irm https://astral.sh/uv/install.ps1 | iex
uvxallows you to run the MCP server without installing the package globally.
pip install patch-tuesday-mcpuvx patch-tuesday-mcpOr install with pip:
pip install patch-tuesday-mcpuvx patch-tuesday-mcp@latestOr with pip:
pip install --upgrade patch-tuesday-mcp
One-click setup: Click the VS Code badge for automatic configuration (requires
uvinstalled) Manual setup: See instructions below for VS Code, Cursor, Claude Code, Copilot CLI, or Claude Desktop
msrc_search – Search and filter Microsoft security updates by keyword, CVE, KB number, month, product, severity, CVSS score, exploited-in-the-wild status, or public disclosure. When no month is given, results default to the most recent release whose Patch Tuesday has already occurred — the upcoming month's pre-release document (early Chromium/out-of-band entries only) is skipped by default and available explicitly via month=. Results are enriched with EPSS scores (FIRST.org 30-day exploitation probability, min_epss=0.5 filter) and CISA KEV catalog status with federal remediation due dates (kev=True filter) — both public, keyless sources. Filter by the parsed CVSS v3.x exposure fields — attack_vector (N/A/L/P), privileges_required (N/L/H), user_interaction (N/R), and scope (U/C) — to isolate, for example, network-reachable zero-click criticals; matching results surface a structured cvss object broken out from the raw vector string. Every CVE detail also includes a references block of ready-to-open links (MSRC update guide, NVD, EPSS API, and the CISA KEV catalog when the CVE is listed). Add include_chain=True to a KB lookup to walk Microsoft-stated supersedence chains (which KBs it replaces, newest → oldest). Pass a list of KB numbers (kb=["5094123", "KB5094127", ...], up to 30) to resolve them all in one call — e.g. a machine's installed-update list as context for a patch report — and get a grouped response with one per-KB entry (found or not-found, each with the same body as a single-KB lookup); every monthly document is still fetched upstream at most once for the whole batch. Add include_known_issues=True to any KB lookup (single or batched) for the Microsoft-confirmed known issues of each update — issue titles, symptoms, workarounds, and the resolving KB when Microsoft names one — scraped best-effort from the KB's public support page with an honest per-KB status: published, none_published (Microsoft publishes no known-issues data for that KB — the norm outside Windows OS updates), or unavailable (retrieval/parse failure, explicitly distinct from "no issues"); it reports what Microsoft has confirmed breaks, not a prediction for your environment. Add include_update_summary=True to a KB lookup for what the update changes — the support page's Summary/Highlights text plus its Improvements bullet list (size-capped, truncated marks a capped block) with the same honest three-way status; both support-page blocks are served by a single fetch and one shared cache record. Add include_guidance=True to a CVE lookup to surface Microsoft-provided mitigations, workarounds, and will-not-fix advisories alongside the vendor-fix KBs. Pass format="markdown" or format="csv" to a monthly/filtered search to get an additive triage briefing — a prioritized executive summary and table (Markdown) or a spreadsheet-ready export with stable columns (CSV) — rendered from the same urgency ranking; the JSON vulnerabilities list is always included. Use force_refresh=True to bypass the in-process caches and re-fetch the MSRC document and EPSS/KEV enrichment for the request, and include_freshness=True to add a freshness block reporting the cache age and TTL of the MSRC document and enrichment data. Search a historical range instead of a single month with months_back=N (the N most recent released months) or start_month/end_month — the response aggregates matching CVEs across the range and adds per-month trend counts; ranges are capped at 12 months and reuse the existing cache/concurrency controls. Set include_stats=True for aggregate counts (by severity, impact, product family, exploited, KEV). Use limit=0 with include_stats=True for a stats-only month overview. Filter on Microsoft's exploitation-likelihood assessment with exploitation_likely=True ("Exploitation More Likely"/"Exploitation Detected"; matches carry an exploitation_assessment field) and on known ransomware campaign use with ransomware=True (from the CISA KEV catalog). Opt into richer rows where you need them: include_references=True adds the MSRC/NVD/EPSS/KEV link block to month/KB/trend list results, include_kev_details=True replaces the boolean KEV flag with the full catalog entry (due date, required action, vendor/product, ransomware use), include_kb_details=True expands KB numbers into full objects with per-KB URL, fixed build, supersedence, sub-type, and restart-required status, and include_temporal=True adds Microsoft's CVSS temporal score to cvss blocks. Filter by weakness class with cwe= (ID or name substring, e.g. cwe="CWE-416"). Use list_months=True to fetch the release catalog (every available month with initial/current release dates — handy for valid month= values and spotting same-month revisions). All new fields are opt-in: the default JSON shape is unchanged.
Scope any search to the products you actually run. Pass product_profile="identity-core" (built-in profiles: identity-core, endpoint, server-infrastructure) or supply ad-hoc products=["Exchange Server", "Windows Server"] / product_families=["Windows", "Azure"]. A vulnerability is kept if it matches any listed product or family. All matching is local — profile contents are never transmitted to MSRC, FIRST.org, CISA, or telemetry.
Override or extend the built-ins by pointing MSRC_PROFILES_PATH at a JSON file:
{
"my-estate": {
"families": ["Windows", "Azure"],
"products": ["Exchange Server", "Microsoft Entra"]
}
}Each entry may set products and/or families (case-insensitive partial matchers). A file entry with the same name as a built-in replaces it. An unknown product_profile, or a missing/invalid MSRC_PROFILES_PATH, returns a clear invalid_input error rather than falling back to a broad, unscoped result.
A portable agent skill — patch-tuesday-triage — teaches an AI agent how to drive msrc_search through the monthly workflow (which searches to run, in what order, how to prioritize). It's plain Markdown and can be deployed independently of this server: copy skills/patch-tuesday-triage/ into your agent's skills directory. See skills/README.md for deployment details.
The server registers an MCP prompt named monthly_triage. MCP clients that support prompts can select it to get a step-by-step analyst workflow built entirely on msrc_search — publicly disclosed zero-days, CISA KEV, exploited, network/no-auth/no-UI criticals, identity-adjacent products, endpoint/Intune, and a briefing. It accepts two optional arguments: product_profile (scope the whole workflow to a watchlist) and month (triage a specific release). No new tools are introduced — the prompt only orchestrates msrc_search calls.
A portable, plain-text copy of this prompt lives under prompts/ so the workflow can be used independently of the server.
Once connected to an MCP client, you can ask questions like:
Monthly overview: "Summarize this month's Patch Tuesday"
Exploited vulnerabilities: "Which Microsoft vulnerabilities are being actively exploited?"
CVE lookup: "What is CVE-2026-41108 and which KB fixes it?"
KB lookup: "What does KB5094123 patch?"
Machine patch report: "Here are the KBs installed on this server: KB5094123, KB5094127, KB5093998 — what do they fix and is anything superseded?"
Product filter: "Show me Critical vulnerabilities affecting Exchange Server this month"
Patch prioritization: "What should I patch first from the June 2026 updates?"
CISA KEV: "Which of this month's CVEs are on the CISA KEV list?"
EPSS: "Show me CVEs with EPSS above 50%"
Exposure filtering: "Which Critical CVEs are network-reachable with no privileges and no user interaction?"
Reference links: "Give me the MSRC, NVD, and EPSS links for CVE-2026-41108"
Mitigations & workarounds: "Are there any mitigations or workarounds for CVE-2026-41108?"
Triage report: "Give me this month's Critical CVEs as a Markdown briefing" (or "…export them as CSV")
Fresh data on demand: "Re-pull this month's updates fresh and tell me how current the data is" (force_refresh=True, include_freshness=True)
Historical trends: "How many HTTP.sys RCE CVEs shipped over the last 6 months?" (query="HTTP.sys", months_back=6)
Supersedence: "Is KB5087538 superseded by anything newer?"
Exploitation forecast: "Which CVEs does Microsoft rate 'Exploitation More Likely' this month?"
Ransomware: "Which of this month's CVEs are used in known ransomware campaigns?"
Deployment planning: "Does KB5094123 require a restart, and which build fixes it?"
Known issues before rollout: "What has Microsoft confirmed breaks in KB5094126, and is there a fix or workaround?" (include_known_issues=True)
What's in an update: "Summarize what KB5099536 changes beyond the security fixes" (include_update_summary=True)
Weakness class: "Show me this month's use-after-free vulnerabilities" (cwe="CWE-416")
Release catalog: "Which Patch Tuesday months are available to query?"
Product watchlist: "Show me this month's Critical CVEs across my estate" (product_profile="identity-core", severity="Critical")
Guided triage: "Walk me through this month's triage for my identity estate" (selects the monthly_triage prompt with product_profile="identity-core")
uvx patch-tuesday-mcpOr if installed with pip:
patch-tuesday-mcpOption 1: One-Click Install (Recommended)
Click the Set up in VS Code badge at the top of this README for automatic configuration (requires uv installed).
Option 2: Manual Configuration
VS Code stores MCP servers in a dedicated mcp.json file — not settings.json. Open the Command Palette (Ctrl+Shift+P on Windows/Linux, Cmd+Shift+P on macOS) and run one of:
MCP: Open User Configuration — edits your user-level mcp.json (available in every workspace)
MCP: Open Workspace Folder Configuration — edits a project-local .vscode/mcp.json
Then add the server. Note that VS Code uses a top-level "servers" key (unlike the "mcpServers" key used by the other clients below):
{
"servers": {
"patch-tuesday": {
"type": "stdio",
"command": "uvx",
"args": ["patch-tuesday-mcp"]
}
}
}Add to your Claude Desktop MCP config:
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
Using uvx (recommended)
{
"mcpServers": {
"patch-tuesday": {
"command": "uvx",
"args": ["patch-tuesday-mcp"]
}
}
}Using installed package
{
"mcpServers": {
"patch-tuesday": {
"command": "patch-tuesday-mcp"
}
}
}Option 1: One-Click Install (Recommended)
cursor://anysphere.cursor-deeplink/mcp/install?name=patch-tuesday-mcp&config=eyJjb21tYW5kIjogInV2eCIsICJhcmdzIjogWyJwYXRjaC10dWVzZGF5LW1jcCJdfQ==Option 2: Manual Configuration
Add to your Cursor MCP config (~/.cursor/mcp.json):
{
"mcpServers": {
"patch-tuesday": {
"command": "uvx",
"args": ["patch-tuesday-mcp"]
}
}
}claude mcp add --transport stdio patch-tuesday -- uvx patch-tuesday-mcpAdd to ~/.copilot/mcp-config.json:
{
"mcpServers": {
"patch-tuesday": {
"type": "stdio",
"command": "uvx",
"args": ["patch-tuesday-mcp"]
}
}
}The server also supports the HTTP transport for remote/shared deployments:
MCP_TRANSPORT=http MCP_PORT=8000 patch-tuesday-mcp
# MCP endpoint: http://localhost:8000/mcpOr with Docker:
docker build -t patch-tuesday-mcp .
docker run -p 8000:8000 patch-tuesday-mcpHTTP-mode environment variables:
Variable | Default | Description |
|
| Set to |
|
| Bind address |
|
| Max request body size, returns 413 above it ( |
|
| Comma-separated allowlist of browser origins. Set an explicit list for public deployments (e.g. |
|
| Max concurrent in-flight connections; uvicorn responds 503 beyond it ( |
|
| Seconds before idle keep-alive connections are closed |
|
| Root log level ( |
|
| Cap on a single MSRC upstream response body (read while streaming, never buffered past the cap) |
|
| Cap on a single EPSS/KEV upstream response body |
|
| Cap on a single support.microsoft.com KB-page body (known-issues / update-summary lookups) |
| unset | Opt-in usage telemetry (requires |
| unset | ISO date ( |
| unset | URL clients should move to |
| unset | Optional ISO date the deprecation was announced (emitted as an RFC 9745 |
| unset | Optional override for the human-readable notice text |
If you are retiring a deployment, set MCP_DEPRECATION_SUNSET and
MCP_DEPRECATION_REPLACEMENT_URL on that deployment only. The server then
reports the migration through four channels:
a deprecation block on every msrc_search response (and on /health)
Deprecation, Sunset, and Link: <url>; rel="successor-version" response
headers (RFC 9745,
RFC 8594,
RFC 5829)
the MCP server instructions shown to clients at connection time
the msrc_search tool description
All of it is descriptive: it states facts about the deployment. It never instructs a connected agent to say or do anything — tool output that issues directives to someone else's model is prompt injection, however benign the payload. Leave the variables unset and behavior is byte-for-byte unchanged; setting only one of the two pair, or a malformed date, is ignored.
HTTP mode also serves GET /health (liveness endpoint) and runs stateless,
so it can scale to multiple replicas behind a load balancer without session
affinity.
The HTTP transport is unauthenticated — msrc_search only reads public
vulnerability data, but an open endpoint is still abusable. Before exposing it
to the internet:
Put it behind an authenticated front door. Terminate TLS and require auth
at a reverse proxy / API gateway (e.g. Azure API Management, an OAuth2/OIDC
proxy such as oauth2-proxy, Cloudflare Access, or your ingress controller's
auth). This server intentionally ships no built-in auth so you can layer your
organization's standard access control in front of it.
Restrict CORS. Set MCP_CORS_ORIGINS to the exact origins of your MCP
clients instead of the permissive * default.
Keep the defaults on. Leave MCP_MAX_BODY_BYTES and
MCP_LIMIT_CONCURRENCY at their defaults (or tighten them) — they are your
first line of defense against oversized payloads and connection exhaustion.
Upstream reads are bounded and redirect-free. Responses from MSRC/EPSS/ KEV are size-capped while streaming and HTTP redirects are never followed, so a misbehaving upstream can't exhaust container memory.
Local stdio usage is unaffected by all of the above; none of this middleware
runs for the default transport.
The container runs on any host that can serve HTTP — Azure Container Apps, Cloud Run, Fly.io, or a plain VM.
Local stdio (the default): no telemetry, ever — there is no code path that sends anything.
The hosted endpoint records minimal usage data to Azure Application Insights (90-day retention): a daily-salted hash of the client IP (raw IPs are never stored; they are only held briefly in memory for abuse protection), request path and timestamp, which tool parameters were used (parameter names only — never your query text or CVE/KB values; only the low-cardinality month and severity values are kept), result counts, latency, and error categories. No cookies, no accounts, no request/response bodies.
Self-hosted HTTP collects nothing unless you set APPLICATIONINSIGHTS_CONNECTION_STRING to your own resource — then the same minimal set flows to your instance instead.
pip install -e ".[dev]"
pytest # offline suite (mocked feeds)
pytest --run-live # also run live smoke tests against the real MSRC / EPSS / KEV APIs
pytest --cov=patch_tuesday_mcp # coverage (CI enforces >= 90%)
ruff check src/ tests/CI (GitHub Actions) runs the offline suite on Python 3.11/3.12/3.14 with a coverage
gate, lints with ruff, builds the container, scans it with Trivy, and produces
an SPDX SBOM on every push/PR. Release builds attest provenance for the
published wheel/sdist. Dependencies are locked in uv.lock (used by the Docker
build via uv sync --locked).
See SECURITY.md for the supported versions, scope, and how to report a vulnerability privately.
MIT
Search Microsoft security updates (Patch Tuesday) from the official MSRC API.
Combines keyword search, CVE/KB lookup, and product/severity/exploitation filtering into a single flexible tool. All filter parameters are optional and can be combined. When no filters are provided, returns the most urgent vulnerabilities from the most recent released Patch Tuesday (CISA-KEV- listed or exploited first, then by EPSS exploitation probability, severity, and CVSS score). The upcoming month's document exists before its Patch Tuesday but only holds early Chromium/third-party and out-of-band entries; it is skipped by default and served only when requested via month=. Results are enriched with EPSS scores (FIRST.org daily exploit prediction, 0-1) and CISA KEV (Known Exploited Vulnerabilities) catalog data when available.
Use this tool to:
Get the latest Patch Tuesday overview (include_stats=True, limit=0)
Browse the most urgent fixes this month (no filters)
Look up a specific CVE with full detail (cve="CVE-2026-41108") -- works across all months, returns KBs, affected products, CVSS, description, FAQs, EPSS score, and KEV status
Find which CVEs a KB article fixes (kb="5094123" or kb="KB5094123") -- scans recent months, or a specific month when combined with month=
Look up many KBs in one call (kb=["5094123", "KB5094127", ...], up to
-- e.g. a machine's installed-update list; returns one grouped per-KB result entry each, with per-KB found/not-found status
Check whether a KB has been superseded by newer patches (kb="5087538", include_chain=True) -- walks Microsoft-stated supersedence links
See what Microsoft has confirmed an update breaks (kb=..., include_known_issues=True) -- known issues from the KB's public support page: symptoms, workarounds, and the resolving update when stated
See what an update changes (kb=..., include_update_summary=True) -- the KB support page's summary and improvements highlights, beyond the security fixes already listed in the MSRC data
Find KEV-listed CVEs this month (kev=True) -- confirmed exploited, with federal remediation due dates
High exploitation probability (min_epss=0.5) -- EPSS >= 50%
Search by keyword (query="Exchange" or query="DNS spoofing")
Filter to a product (product="Windows Server 2022") -- partial match
Filter to a product watchlist (product_profile="identity-core", or products=["Exchange Server", "Windows Server"] / product_families=["Windows", "Azure"]) -- keeps vulns matching any listed product or family; matching is local and profiles never leave the host
Filter by severity (severity="Critical") -- Critical/Important/Moderate/Low
Find actively exploited vulnerabilities (exploited=True)
Find publicly disclosed zero-days (publicly_disclosed=True)
Filter by CVSS score (min_cvss=8.0)
Look at a past month (month="2026-Apr" or month="2026-04")
Combine filters (product="Exchange" + severity="Critical" + month="2026-05")
Search a historical range (query="HTTP.sys" + months_back=6, or start_month="2026-Jan" + end_month="2026-Jun") -- aggregates matching CVEs across released months with per-month trend counts
Paginate with offset (offset=10, limit=10 for page 2)
Args: query: Optional keyword; case-insensitive match across CVE ID, title, description, component tag, and affected product names. cve: Optional CVE ID (e.g. "CVE-2026-41108"). Fast path: ignores other filters and returns full detail for that single CVE, searching across all months automatically. kb: Optional KB article number (e.g. "5094123" or "KB5094123") or a list of up to 30 of them for a batched lookup. Fast path: returns the CVEs fixed by that KB, scanning the most recent months (up to 6), or only the given month when month= is also set. A list input returns a grouped response instead: a results array with one entry per KB ("kb", "found", and on success the same body as a single-KB lookup; on a miss "error"/"error_kind"), deduplicated, order preserved. limit/offset apply per KB; other filters are ignored. Accepts numeric KB ids only (any malformed list entry fails the whole call); kb_articles in results may also contain non-KB vendor-fix labels such as "Release Notes", which cannot be looked up here. month: Optional monthly release to search, formatted "2026-Apr" or "2026-04". Defaults to the most recent release whose Patch Tuesday (second Tuesday of the month) has already occurred; pass the upcoming month explicitly to see its pre-release entries. Combined with kb=, restricts the KB lookup to that month. product: Optional product name filter (case-insensitive partial match against affected product names, e.g. "Windows Server 2022"). product_profile: Optional named watchlist (e.g. "identity-core") that expands locally into product/family matchers. Built-in profiles can be overridden/extended via a JSON file at MSRC_PROFILES_PATH. An unknown name returns an invalid_input error listing the available profiles. Profile contents are matched locally and never sent to MSRC, FIRST.org, CISA, or telemetry. products: Optional list of product-name partial matchers. A vulnerability is kept if any entry matches one of its affected products. product_families: Optional list of product-family partial matchers. A vulnerability is kept if any entry matches one of its product families. Combined with products/product_profile as a union (match any listed product OR family). severity: Optional maximum-severity filter. Valid values: Critical, Important, Moderate, Low. exploited: Optional filter for vulnerabilities known to be exploited in the wild (True) or not (False), per Microsoft's assessment. publicly_disclosed: Optional filter for publicly disclosed vulnerabilities. kev: Optional filter for CVEs on (True) or off (False) the CISA Known Exploited Vulnerabilities catalog. ransomware: Optional filter for CVEs whose CISA KEV entry reports known ransomware campaign use (True) or not / not on KEV (False). exploitation_likely: Optional filter on Microsoft's latest-release exploitation assessment: True matches "Exploitation More Likely" or "Exploitation Detected"; False matches everything else (including unassessed entries). Matching results include an exploitation_assessment field. cwe: Optional weakness filter; case-insensitive substring match against CWE entries (e.g. "CWE-416" or "use after free"). Matching results include their cwe list. min_epss: Optional minimum EPSS score (0-1), the probability of exploitation in the next 30 days (e.g. 0.5 for >= 50%). min_cvss: Optional minimum CVSS base score (0-10). attack_vector: Optional CVSS attack-vector filter, one of N (network), A (adjacent), L (local), P (physical). Matches the parsed CVSS v3.x vector; entries without a parseable vector are excluded. privileges_required: Optional CVSS privileges-required filter, one of N (none), L (low), H (high). user_interaction: Optional CVSS user-interaction filter, one of N (none), R (required). scope: Optional CVSS scope filter, one of U (unchanged), C (changed). include_chain: When True together with kb=, adds a supersedence_chain showing which KBs this KB replaces (newest to oldest), walked from Microsoft-stated supersedence links. Ignored without kb=. include_guidance: When True together with cve=, adds a guidance list to the CVE detail output with any Microsoft-provided mitigations, workarounds, and will-not-fix advisories (type/description/url). Omitted by default to keep responses lean. Ignored without cve=. include_references: When True, adds deterministic reference links (MSRC, NVD, EPSS, and KEV when listed) to each result row in month/KB/trend lists. CVE detail lookups always include them. include_kb_details: When True, kb_articles entries become full objects (per-KB url, fixed_build, supersedence, sub_type, and restart_required) instead of bare KB numbers; on cve= lookups it additionally adds restart_required to each KB entry. include_kev_details: When True, KEV-listed rows carry the full KEV entry (due_date, ransomware_use, required_action, vendor_project, product, vulnerability_name) instead of a boolean flag; on cve= lookups it extends the kev block with the extra catalog fields. include_known_issues: When True together with kb=, adds a known_issues block per KB with the issues Microsoft has publicly confirmed for that update (title, symptoms, workaround, and the resolving KB when stated), scraped best-effort from the KB's support page on support.microsoft.com. This reports what Microsoft has confirmed breaks -- it does not predict behavior in a specific environment. The block's status field is honest about coverage: "published" (issues listed), "none_published" (Microsoft publishes no known-issues data for this KB -- the norm for most non-Windows products; Windows cumulative/preview updates are the main source), or "unavailable" (the page could not be fetched or parsed; NOT the same as no issues). Attached even when the KB is not found in MSRC security releases (e.g. preview-only updates). Ignored without kb=. include_update_summary: When True together with kb=, adds an update_summary block per KB describing what the update changes, scraped best-effort from the same support.microsoft.com KB page: the page's Summary/Highlights text plus its Improvements bullet list (size-capped; truncated=True marks a capped block). The status field is honest like known_issues: "published" (a summary exists), "none_published" (Microsoft publishes no summary section for this KB), or "unavailable" (the page could not be fetched or parsed; NOT the same as no summary). Attached even when the KB is not found in MSRC security releases; combines freely with include_known_issues -- both blocks are served by a single fetch of the same page. Ignored without kb=. include_temporal: When True, cvss blocks gain the CVSS temporal score Microsoft publishes (exploit-code maturity adjusted). Applies to cve= detail and to list rows that carry a cvss block. list_months: When True, ignore other filters and return the catalog of available monthly releases (id, title, initial/current release dates, newest first) — useful for discovering valid month= values and spotting same-month revisions. format: Output format for a monthly/filtered search: "json" (default, most complete), "markdown", or "csv". "markdown" adds a prioritized triage briefing (executive summary + table) under a markdown key; "csv" adds a spreadsheet-ready table under a csv key plus a columns list. The JSON vulnerabilities list is always included. Ignored for cve=/kb= fast-path lookups. report: Optional report profile for format="markdown"/"csv". Currently only "triage" (the default rendering) is supported; reserved for future report shapes. force_refresh: When True, bypass the in-process caches for this request and re-fetch the MSRC document and EPSS/KEV enrichment from source. Use to pick up a same-day MSRC revision or fresh EPSS/KEV data. Only the data needed for this request is refreshed; unrelated cached months are left intact. include_freshness: When True (or when force_refresh is used), add a freshness block to the response reporting the cache age and TTL of the MSRC document and the EPSS/KEV enrichment data. months_back: Optional historical-trend control; search the N most recent released months (N >= 1) instead of a single month, aggregating matches with per-month counts. Mutually exclusive with start_month/end_month. Capped at 12 months per request. start_month: Optional start of a historical-trend range (e.g. "2026-Jan" or "2026-01"), inclusive. When end_month is omitted the range runs through the latest released month. Capped at 12 months. end_month: Optional end of a historical-trend range (inclusive); requires start_month (or months_back). Pre-release months are excluded. limit: Maximum number of results to return (default: 10, max: 100). Set to 0 with include_stats=True for a stats-only month overview. offset: Number of results to skip for pagination (default: 0). include_stats: When True, includes aggregate counts (by severity, impact, product family, exploited, KEV, publicly disclosed) for the filtered result set.
Returns: Dictionary with: - month: Release ID (e.g. "2026-Jun") and title/release date - total_found: Number of vulnerabilities matching the filters - vulnerabilities: List of compact vulnerability summaries (up to limit) with epss_score and kev flag when available; full detail (epss_percentile, KEV due dates) returned for cve= lookups - filters_applied: Summary of which filters were used - stats: (only when include_stats=True) aggregate counts - supersedence_chain / chain_complete: (only for kb= lookups with include_chain=True) the walked chain, newest to oldest - known_issues: (only for kb= lookups with include_known_issues=True) per-KB block with status ("published" / "none_published" / "unavailable"), an issues list (title, symptoms, workaround, resolution, resolved_by) when published, a note otherwise, and the source_url of the Microsoft support page - update_summary: (only for kb= lookups with include_update_summary=True) per-KB block with status ("published" / "none_published" / "unavailable"), the page title, a summary string and an improvements list of what the update changes when published (truncated=True when size caps trimmed content), a note otherwise, and the source_url of the Microsoft support page - total_kbs / results: (only when kb= is a list) grouped batch output; results holds one entry per KB with kb, found, and either the single-KB response body or a per-KB error/error_kind, while the top-level total_found sums across all KBs - guidance: (only for cve= lookups with include_guidance=True) list of mitigation/workaround/will-not-fix advisories, when Microsoft provides them - format / markdown / csv / columns: (only when format="markdown" or "csv") the chosen format plus the rendered triage view; csv also carries the stable column-name list - freshness: (only with include_freshness=True or force_refresh=True) cache age/TTL for the MSRC document and EPSS/KEV enrichment - range / months_searched / trend: (only for historical-trend searches via months_back or start_month/end_month) the resolved month range, the number of months searched, and per-month aggregate counts (total, by_severity, exploited, publicly_disclosed, kev) - available_months: (only with list_months=True) the release catalog, newest first - error / error_kind: (only on failure) a message plus a category (invalid_input, not_found, upstream, internal) - note: (when relevant) explains month selection, e.g. that a newer pre-Patch-Tuesday document was skipped, or (with release_status="pre-patch-tuesday") that the requested month has not had its Patch Tuesday yet
| Name | Required | Description | Default |
|---|---|---|---|
| kb | No | ||
| cve | No | ||
| cwe | No | ||
| kev | No | ||
| limit | No | ||
| month | No | ||
| query | No | ||
| scope | No | ||
| format | No | json | |
| offset | No | ||
| report | No | ||
| product | No | ||
| min_cvss | No | ||
| min_epss | No | ||
| products | No | ||
| severity | No | ||
| end_month | No | ||
| exploited | No | ||
| ransomware | No | ||
| list_months | No | ||
| months_back | No | ||
| start_month | No | ||
| attack_vector | No | ||
| force_refresh | No | ||
| include_chain | No | ||
| include_stats | No | ||
| product_profile | No | ||
| include_guidance | No | ||
| include_temporal | No | ||
| product_families | No | ||
| user_interaction | No | ||
| include_freshness | No | ||
| include_kb_details | No | ||
| include_references | No | ||
| publicly_disclosed | No | ||
| exploitation_likely | No | ||
| include_kev_details | No | ||
| privileges_required | No | ||
| include_known_issues | No | ||
| include_update_summary | No |
| Name | Required | Description |
|---|---|---|
No output parameters | ||
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already indicate read-only, idempotent, and open-world traits. The description adds substantial behavioral context: explains month-selection logic, cache behavior, best-effort scraping for known issues, EPSS/KEV enrichment, and error kinds. No contradictions 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?
The description is very long, but well-structured with a summary, use-case list, parameter descriptions, and return section. While every sentence adds value, the length may hinder quick scanning for an AI agent, so conciseness is average.
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 40 parameters, no required params, 0% schema coverage, and a detailed output schema described, the description is exhaustive. It covers all parameter behaviors, return values, error states, edge cases (e.g., pre-release months, cache refresh), and enrichment details. No 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?
With 0% schema description coverage, the description fully compensates by explaining each of the 40 parameters in detail, including constraints, fast paths, default behavior, and usage notes (e.g., kb parameter supports list, month formats, product_profile expansion). This is essential for correct usage.
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 searches Microsoft security updates from the MSRC API, combining keyword search, CVE/KB lookup, and filtering. It explicitly distinguishes the tool's purpose with a specific verb and resource, and the lack of sibling tools does not detract from clarity.
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?
A bulleted list of use cases provides explicit guidance on when to use the tool, covering scenarios like getting Patch Tuesday overview, looking up CVEs, finding KB fixes, checking supersedence, known issues, and more. This is exemplary for an AI agent.
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.
v0.9.1msrc_search1 field changedInput schema / properties / include_update_summaryAdded value: +{
+ "default": false,
+ "type": "boolean"
+}v0.8.0msrc_search1 field changedInput schema / properties / include_known_issuesAdded value: +{
+ "default": false,
+ "type": "boolean"
+}v0.1.0msrc_searchThere is only one tool, so there is no risk of confusion between tools. The tool's purpose is clearly defined.
With a single tool, naming consistency is not an issue. The name 'msrc_search' is descriptive and follows a clear convention.
Having only one tool for the entire server is borderline. While the tool is comprehensive, bundling all functionality (search, CVE/KB lookup, trends) into one makes it complex and less discoverable. A small set of separate tools would be more appropriate.
The single tool covers all expected operations for the MSRC domain: searching, filtering, CVE/KB detail, historical trends, pagination, stats, and enrichment. There are no apparent gaps in the tool surface.
Unclaimed servers have limited discoverability.
If you are the server author, to access and configure the admin panel.
CVE & vulnerability search: 365k+ CVEs/NotCVEs, CVSS, EPSS, CISA KEV, exploits, patches, versions.
Threat intel + your scans/findings/Shield posture. CVE, EPSS, KEV, package vuln lookup, DAST.
CVE search, vulnerability database, EPSS exploit prediction, KEV, IP reputation & threat feed.
CVE intelligence: exploitation (KEV/EPSS), detection coverage, fixed versions. All tools keyless.
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/jonnybottles/patch-tuesday-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server
