Blobfish MCP
Any OpenAPI spec. Zero config. Claude-ready. An MCP server that turns any REST API into Claude-callable tools at runtime. Point it at an OpenAPI URL, Swagger spec, or Postman collection and Claude can immediately call every endpoint, authenticated and live.
Stack: Node.js, MCP SDK, OpenAPI 3.x, Swagger 2.0, Postman
GitHub · Live
How it works: Blobfish starts with 15 meta-tools Claude can always call. When Claude calls load_api or discover_api, Blobfish parses the spec and fires a tools/list_changed notification. New tools appear in Claude's context immediately, with no restart required.
discover_api: Auto-finds a spec from just a domain by probing 15 common paths. You give it a URL, it finds the OpenAPI spec itself.
fetch_all: Auto-paginates any endpoint. Detects and follows Link headers (GitHub/Stripe style), cursor fields, has_more patterns, and total/offset/limit. One call, all pages.
run_workflow: Chains multi-step API calls with {{ template }} syntax, foreach loops over response arrays, and run_if conditionals. Build a pipeline that fetches a user, iterates their repos, and filters by criteria, all in one Claude prompt.
get_last_request_log: Shows the exact URL and request body of the last N calls. When a 400 comes back, Claude can call this, see what it sent, and self-correct.
blobfish.json: Pre-configure APIs to load at startup with per-API auth (bearer, API key, or basic auth), timeout, and tag filtering. Environment variable values like ${MY_API_TOKEN} are interpolated at startup.
Mock mode: Load any spec in mock mode to get example responses without real HTTP calls, generated from the OpenAPI example fields. Useful for testing or demos without API keys.
Supported formats: OpenAPI 3.x (JSON and YAML), Swagger 2.0 (JSON and YAML), Postman Collections v2.1, and local files.
Distribution: npx blobfish-mcp <url> runs it directly without installing. npm run setup auto-detects the Claude Desktop config path and writes the entry. Compatible with Cursor, Zed, and Smithery.
swayam.codes