OpenAPI Search
A small, open directory of OpenAPI descriptions for popular APIs, with an MCP server that turns any of them, or any spec URL, into a short plain-language summary an AI model can read before writing calls.
- What it is
- Agent tools and integrations
- Used for
- Web APIs
- Type
- Open source
- Made by
- Jan Wilmake
- License
- MIT (MCP server); the directory repository has no license file
- Facts checked
- 13 October 2026
What it is
OpenAPI Search is a one-person open-source project by Jan Wilmake. Its core is a hand-curated providers.json that maps about 66 short ids, such as stripe, github, supabase or hackernews, to the URL of each API’s OpenAPI file, and https://openapisearch.com/redirect/<id> sends you to that file. The site also serves the list as Markdown and llms.txt, so a model can read the whole directory in one request.
The useful part for agents is the companion MCP server, a Cloudflare Worker reachable at https://openapi-mcp.openapisearch.com/mcp. It has two tools. getApiOverview takes a directory id or a raw spec URL, fetches the JSON or YAML (converting old Swagger 2.0 files through swagger.io’s converter), and returns a plain-language list of the endpoints. getApiOperation then returns the detail for one operation by id or route. The same summaries are served over plain HTTP at oapis.org, so the model reads a short summary instead of the full spec.
It is a side project rather than a product: there is no account, pricing or service-level promise, the directory covers dozens of APIs rather than thousands, and its own README says the author chose to focus on the quality of a few great specs over quantity.
Why teams pick it
- Free and open: the MCP server is MIT-licensed and runs as a single Cloudflare Worker you can deploy yourself
- Works with any OpenAPI or Swagger URL, not only the curated ids
- Small context cost: a model gets a summary and then one operation at a time instead of the whole spec
Watch out for
- A small, hand-maintained list of about 66 APIs, and the directory repository has no license file
- Run by one person with no stated uptime or support; the summaries depend on oapis.org and swagger.io’s public converter
- It only describes APIs: it does no authentication or calls, so the agent still needs keys and an HTTP tool to act
Alternatives
What all agent tools and integrations have in common →Compare side by side →
ComposioCommercialA hosted tool layer for AI agents: one session per end user gives the model a few meta tools to search, authorize and run actions across more than 1,500 app toolkits, with Composio holding and refreshing the OAuth tokens.
GlamaCommercialThe largest public directory of MCP servers, which builds and runs each open-source server in a Firecracker sandbox to record its tools, score their descriptions and flag risky behavior, plus hosting and a gateway for the servers you use.
Model Context Protocol (MCP)Open sourceAn open protocol, built on JSON-RPC, that lets an AI application discover and call tools, read resources and use prompt templates from any server that speaks it, so one integration works across Claude, ChatGPT, Cursor, VS Code and other clients.
OneRenamedCommercialA hosted connection layer for AI agents, formerly Pica: you connect each app once in your own account, and every agent reaches it through one MCP server or CLI, with per-action scopes enforced and logged on each call.
SpeakeasyCommercialGenerates typed SDKs, Terraform providers, CLIs and MCP servers from an OpenAPI document, and now also sells a hosted MCP platform and an enterprise MCP gateway that governs which agents may call which tools.How it fits with other technologies
Read an API the way the MCP server serves it · bash
# The directory, as Markdown with one line per API
curl -s https://openapisearch.com/llms.txt
# A plain-language overview of one API (the getApiOverview tool returns the same)
curl -s https://oapis.org/llms.txt/hackernews
# The raw OpenAPI file behind a directory id
curl -sL https://openapisearch.com/redirect/hackernews
# MCP clients connect to: https://openapi-mcp.openapisearch.com/mcpSee it in a real stack
No architecture uses OpenAPI Search yet. Drop it into the playground and wire it up yourself.Open playground →Where it came from2024–2026
Jan Wilmake started the directory in January 2025 because, in their words, OpenAPI descriptions “are effectively closed if there isn’t an accessible way to find and use them”: commercial marketplaces such as RapidAPI dominated search results, and the best open directory, APIs.guru, listed only a fraction of public APIs. The goal was a registry easy for agents to navigate rather than a docs site.
In March 2025 Wilmake picked the project back up and relaunched it with a narrower aim, quality over quantity, and paired it with an MCP server so Claude and Cursor could use it. They posted it to Hacker News on 22 March 2025 as “Make APIs accessible for AI codegen”, warning readers to expect bugs. Later updates added an llms.txt file and a skills endpoint (January 2026) and a handful of fixes.
Why it’s built this way
- Summaries in simple language instead of raw specs
- Large OpenAPI files overflow a model’s context, so the MCP server first identifies the API, then returns an overview, then the details of only the endpoints the model needs.
- Quality of a few specs over quantity
- Wilmake’s March 2025 changelog notes that people want to see their own APIs and aren’t interested in 99% of the others, so the project curates a short list instead of crawling for thousands.
- Stay inside the OpenAPI standard
- The README lists custom, non-standard logic as a non-goal: improvements should be a layer on top of plain OpenAPI files.
Timeline
- 2024
openapi-mcp-server repository created in December
- 2025
First version of the directory in January
- 2025
Relaunched with the MCP server; Show HN on 22 March
- 2026
llms.txt and a skills endpoint added in January