DocStar MCP
Hosted MCP server from DocStar (docstar.io), API documentation for people and AI agents: agents search API docs and call the API as tools, with OAuth sign-in.
ドキュメント
A hosted MCP server for your API.
MCP server
DocStar hosts an MCP server for every API you document. Agents search your docs, read the exact contract and make real calls, signed in with each user's own access through OAuth. No server to build, host or keep in sync.
Copy prompt: paste this into your agent:
Use DocStar to let my users use my app from their AI assistants.
Follow https://docstar.io/mcp/quickstart.md, document my app's user-facing APIs from my latest code, and set up MCP.
Works with: Claude Code, Claude Desktop, ChatGPT, Cursor, VS Code, any MCP client
connect your agent:
$ claude mcp add --transport http acme https://mcp.acme.com
✓ Connected to acme · 3 tools
search_acme_docs
get_acme_api_docs
call_any_api_of_acme
From your code to an MCP server, without writing one.
How it works
- Your agent documents your API. Paste the prompt into your coding agent. It reads your routes and creates the API docs in DocStar.
- DocStar creates the MCP server. Every API docs project gets a hosted MCP server, named after your API, on a DocStar URL or your own domain.
- Any agent can use your API. Your users connect their agents in one step. The tools read your published docs, so they're always current.
What agents can do with your API.
Tools
Every MCP server comes with tools to find, read and call your API. Add your own for the jobs that matter most.
Find the right endpoint
find
Agents search your docs in plain language and land on the right endpoint, guide or example.
search_acme_docs “send a WhatsApp template”
Read the exact contract
read
They get methods, URLs, parameters, request bodies, responses and examples, straight from your published docs.
get_acme_api_docs → refunds module
Make real calls
act
Agents call your API for the user, signed in with the user's own access, and confirm before anything that changes data.
POST /v1/refunds → 201 Created
Add your own tools
your tools
Turn key endpoints into dedicated tools, like create_refund, so agents use them directly. Write instructions that tell agents how your API works.
create_refund · get_order · list_payments
Works with every MCP client.
Connect
Claude Code
Add it from the terminal.
claude mcp add --transport http acme https://mcp.acme.com
Cursor and VS Code
Add it to your MCP settings.
{ "mcpServers": { "acme": { "url": "https://mcp.acme.com" } } }
ChatGPT and Claude Desktop
Add a custom connector and paste the URL. Users sign in with OAuth when the tools need it.
Run it where you need it.
Hosted or local
Hosted MCP server
hosted by DocStar · HTTP
Paste one URL into your client. Find, read and call tools plus your own, with OAuth sign-in for each user. Best for sharing your API with every customer's agents.
docstar-cli mcp
on your machine · stdio
Install your API with docstar-cli, then run it as a local MCP server with one tool per endpoint. Best for scripting, testing and power users.
Hosted MCP server questions.
FAQ
What is an MCP server for an API?
An MCP (Model Context Protocol) server gives AI agents tools they can call. For an API, those tools let an agent find the right endpoint, read its contract and make the request, instead of guessing from copied docs.
Do I need to build or host anything?
No. DocStar creates and hosts the MCP server for each API docs project. You publish your docs; the server follows them.
How do the tools stay up to date?
The tools read your published docs. Republish and every connected agent sees the new contract.
How do users authenticate?
The hosted server supports OAuth sign-in, so each user's agent calls your API with that user's own access.
Can it run on my own domain?
Yes. The MCP server can run on your domain, for example mcp.acme.com, with DNS and certificate set up from DocStar.
Can agents change data through it?
Only through the write tools, which are separate from the read tools, so agents can ask the user before anything that changes data. Read-only GET requests have their own tool.
Can I give important endpoints their own tools?
Yes. Turn key endpoints into named tools, like create_refund, edit the tool descriptions and the server's skill, and add instructions that tell agents how your API works.
Hosted server or docstar-cli mcp: which should I use?
The hosted server is for sharing your API with every customer's agents, with OAuth sign-in. docstar-cli mcp runs on one machine over stdio with one tool per endpoint, for scripting and testing.
Built from the same docs.
Related
API docs
A public API reference with a try-it playground, guides, llms.txt and markdown for every page.
docstar-cli
Every published endpoint becomes a terminal command, and the CLI runs as a local MCP server.
Agent analytics
See which people and agents read your docs, what they ask and what they couldn't find.
Your API, ready for every agent.
One prompt. Your docs, MCP server and CLI, live.
Copy prompt: paste this into your agent:
Use DocStar to let my users use my app from their AI assistants.
Follow https://docstar.io/mcp/quickstart.md, document my app's user-facing APIs from my latest code, and set up MCP.