Osv Advisory MCP Server README

Overview

Vulnerability data from OSV.dev, the open-source vulnerability database. Query a package version or Git commit, batch-audit a dependency list or SBOM, and fetch complete advisory records with CVSS severity, CVE aliases, and affected version ranges.

Tools

Capability reference

osv_query_package _tool

• Provide name, ecosystem (case-sensitive exact match), and version, or a purl such as pkg:pypi/jinja2@3.1.4. A PURL needs exactly one version: embedded in the URL or supplied separately. Versions are exact, never ranges
• Unsupported ecosystem-specific constraints fail as invalid_version; npm/crates.io exact pins with = or == normalize to the installed version
• Each advisory carries CVE aliases, severityLabel with its severitySource, the queried package's fixedVersions, affectedRanges, and cweIds
• truncated: true means OSV had more pages than OSV_QUERY_MAX_PAGES follows, so an empty vulns list is not a clean result; an unrecognized ecosystem fails as invalid_ecosystem

osv_query_batch _tool

• 1–1000 {name, ecosystem, version} tuples or {purl, version?} entries per call; mix either form, and results[i] matches packages[i]
• Each row carries vulnerable, vulnCount, vulns, truncated, and error; a failure lands in the affected rows' error instead of failing the call
• summary totals the scan (vulnerableCount, cleanCount, truncatedCount, errorCount, totalVulns, worstSeverity), and truncated rows never count as clean. The whole scan has a 50-second deadline
• vulnerableRows, errorRows, and truncatedRows in summary retain original zero-based indices even when the preview omits those rows. Lists can overlap; retrieve a row with /results/<index>

osv_query_commit _tool

• One full commit: 40 hexadecimal characters for SHA-1 or 64 for SHA-256; either letter case is accepted, abbreviations are rejected
• Advisories retain record-wide severity, aliases, and GIT ranges with repository URLs and ordered events. No package-specific fixedVersions are inferred
• truncated: true means more OSV pages remain; an empty truncated result is not confirmed clean

osv_get_vulnerability _tool

• One exact advisory ID from any OSV source (GHSA-, PYSEC-, RUSTSEC-, GO-, DSA-, USN-, CVE-, …), matched case-sensitively
• Full record: details, aliases, every affected package with its ranges and fixes, severityLabel, cweIds, and references
• An unknown ID fails as vulnerability_not_found; withdrawn is set only on retracted advisories

osv_list_ecosystems _tool

• No input; returns the ecosystems strings the query tools accept, plus a note with the date the list was last checked against OSV.dev
• Strings are case-sensitive exact matches: PyPI works, pypi fails

Features

Built on @cyanheads/mcp-ts-core: stdio and Streamable HTTP transports, pluggable auth (none / jwt / oauth), swappable storage (in-memory, filesystem, Supabase, Cloudflare KV/R2/D1), structured logging with optional OpenTelemetry tracing.

OSV-specific:

• No API key required — OSV.dev is public and keyless, with no published rate limit
• osv_query_batch finds matching advisory IDs through OSV's /v1/querybatch, then fetches each unique advisory once, so batch rows carry the full records (aliases included) the batch endpoint omits; requests are capped by OSV_BATCH_CONCURRENCY
• osv_list_ecosystems is checked against the live OSV schema and API with bun run check:ecosystems

Agent-friendly output:

• Truncation is never read as clean — truncated flags incomplete OSV pagination, and truncated batch rows are left out of cleanCount
• aliases (CVE IDs) on every advisory chain to nist-nvd-mcp-server for CVSS base scores and CISA KEV status
• Responses are capped at 24,000 bytes. Start retrieval by repeating the original input with page: overflow.page. Continue the same path with offset: page.nextOffset and revision: page.revision until nextOffset is null. Fetch each entries[].deferred.path separately at offset zero without a revision; never copy the whole output page object into the input
• Advisory text in content[] escapes HTML and unsafe-link constructs while preserving HTTP(S) links; caller coordinates render literally. structuredContent keeps every OSV string verbatim

Getting started

Use the public hosted instance with Claude Code:

claude mcp add --transport http osv-advisory-mcp-server https://osv-advisory.caseyjhand.com/mcp

Or run locally through npm:

claude mcp add osv-advisory-mcp-server -- npx -y @cyanheads/osv-advisory-mcp-server

For other MCP clients, add this to the client configuration:

{
  "mcpServers": {
    "osv-advisory-mcp-server": {
      "command": "npx",
      "args": ["-y", "@cyanheads/osv-advisory-mcp-server"]
    }
  }
}

No API key is required. For Docker or a source build, see Running the server.

Configuration

All configuration is validated at startup, and every variable is optional.

See .env.example for the full list of optional overrides.

Running the server

Local development

Requires Bun v1.4.0 or higher (or Node.js v24+).

git clone https://github.com/cyanheads/osv-advisory-mcp-server.git
cd osv-advisory-mcp-server
bun install
cp .env.example .env

• Build and run:

  shCopy

  # One-time build
  bun run rebuild
  
  # Run the built server
  bun run start:stdio
  # or
  bun run start:http
  

• Run checks and tests:

  shCopy

  bun run devcheck   # Lint, format, typecheck, security
  bun run test       # Vitest test suite
  bun run lint:mcp   # Validate MCP definitions against spec
  bun run check:ecosystems  # Compare osv_list_ecosystems with the live OSV schema and API
  

Docker

docker run --rm -p 3010:3010 ghcr.io/cyanheads/osv-advisory-mcp-server:latest

The image defaults to HTTP transport, stateless session mode, and logs to /var/log/osv-advisory-mcp-server. For stdio, use docker run -i --rm -e MCP_TRANSPORT_TYPE=stdio ghcr.io/cyanheads/osv-advisory-mcp-server:latest. Build locally with docker build -t osv-advisory-mcp-server .; OpenTelemetry peers are installed by default — use --build-arg OTEL_ENABLED=false to omit them.

Project structure

Development guide

See CLAUDE.md/AGENTS.md for development guidelines and architectural rules. The short version:

• Handlers throw, framework catches — no try/catch in tool logic
• Use ctx.log for request-scoped logging, ctx.enrich for response context, and ctx.signal for cancellable OSV requests
• Register new tools via the barrel in src/mcp-server/tools/definitions/index.ts
• Wrap external API calls: validate raw → normalize to domain type → return output schema; never fabricate missing fields

Contributing

Issues are welcome. Run checks and tests before submitting:

bun run devcheck
bun run test

License

Apache-2.0 — see LICENSE for details.