Give your AI agent real tools.

YourMCP is a Model Context Protocol server you can ship today. One service registry, exposed identically over CLI, MCP, and HTTP, so any agent that speaks MCP can call it.

View on GitHub Read the docs
Claude

Triage my inbox and draft replies.

I triaged your inbox. Here's the curated view.

gmail_curate_inbox ✓
Curated inbox
123
  • D
    Re: Q3 contract - countersignature 0.86
    Draft × Unread awaiting your reply
    Dana Whitfield
    Looks good on our end. Just need your signature on page 4 before EOD.
    2 days ago
  • P
    Design review for v2 dashboard 0.61
    Unread mentions you
    Priya Nair
    Pushing our review an hour, calendar invite updated.
    5 hours ago
  • S
    Your invoice is ready 0.42
    has attachment
    [email protected]
    Your monthly statement is attached. Total due $1,204.00.
    1 day ago
Inbox
Re: Q3 contract - countersignature
D Dana Whitfield <[email protected]>

Looks good on our end. Just need your signature on page 4 before EOD - once that's in we can countersign and get this filed today.

Draft a researched reply to the Q3 contract thread from Dana.

gmail_reply_to_thread ✓

Researched the thread and drafted a reply, saved to your inbox.

Inbox Draft ready
Re: Q3 contract - countersignature
D Dana Whitfield Just need your signature on page 4… 2d
To [email protected]
Draft saved
Reply to Claude…

Works with every MCP client

  • Claude
  • Codex
  • Cursor
  • ChatGPT
  • VS Code
  • OpenClaw
  • Goose

One tool. Three transports.

Write a service once and call it identically from the CLI, any MCP client, or plain HTTP. Same inputs, same outputs, zero duplicated logic. Pick yours to get set up.

01

Add it to your client

The server runs over streamable HTTP, so onboarding is just its URL. Pick your client: one click where deep links are supported, copy-and-paste everywhere else.

https://your-app.up.railway.app/mcp
Add to Cursor

Opens Cursor and adds the server. Not working? Copy the URL above and add it under Settings → MCP.

02

Call a tool

Your agent discovers the tools automatically and calls them with typed inputs. The same service answers two ways:

The default: a pure service returns its typed output model. FastMCP derives the outputSchema, so the CLI, API, and MCP all behave identically.

jsonc 
// client → server  ·  JSON-RPC over streamable HTTP
{
  "method": "tools/call",
  "params": {
    "name": "gmail_curate_inbox",
    "arguments": { "limit": 3 }
  }
}

// server → client  ·  structured output
{
  "content": [{ "type": "text", "text": "3 threads ranked. Top: Q3 contract redlines" }],
  "structuredContent": {
    "threads": [
      { "subject": "Re: Q3 contract redlines", "importance_score": 0.86, "has_draft": true },
      { "subject": "Design review for v2 dashboard", "importance_score": 0.61, "has_draft": false }
    ]
  }
}

One codebase, every surface

Write a tool once. Ship it to agents, scripts, and services without rewrites.

  1. 01

    Three transports, zero duplication

    Every tool is a pure function in a shared registry, exposed identically over CLI, MCP, and HTTP. Behavior never drifts between interfaces.

  2. 02

    Headless or interactive

    Return data for autonomous agents, or opt into enhanced tools that elicit input, attach media, and render sandboxed UI dashboards.

Trusted by builders

“We had a production MCP server in front of our agents the same afternoon. The shared registry meant our CLI and API just worked too.”
Placeholder NameStaff Engineer, Placeholder Co.
“The typed schemas are the killer feature. Our agent stopped guessing argument shapes overnight.”
Placeholder NameFounder, Placeholder AI

Ask AI about this

Have your assistant explain the template, compare it, or walk you through deploying it.

Ask in ChatGPT Ask in Perplexity Ask in Claude

Frequently asked questions

Which MCP clients are supported?

Any client that speaks the Model Context Protocol: Claude Desktop, Claude Code, Cursor, Cline, VS Code, Windsurf, and more. The server exposes a standard tool/resource surface.

stdio or streamable HTTP?

Both. Streamable HTTP is the primary transport (mounted at /mcp alongside the HTTP API in one process), and stdio is available for local/dev use.

How does authentication work?

The MCP mount supports OAuth 2.1 as a resource server, sharing auth and CORS with the HTTP API. You can also run it unauthenticated for local development.

Do I need to install anything to use it?

No. Because the server runs over streamable HTTP, connecting is just pasting its URL into your agent client. No local install, runtime, or download required. (Self-hosting the server is a separate, optional step.)

Does it work on mobile?

Yes, anywhere your agent runs. Since it's a remote HTTP server with nothing to install locally, it works in any agent app that has a mobile app, including the Claude and ChatGPT mobile apps.

Can I self-host?

Yes. The whole thing is open source and ships with a Dockerfile and Railway config. Deploy it anywhere that runs a container.

What about my existing CLI / API?

They share the same service registry. Add a tool once and it's available over CLI, MCP, and HTTP simultaneously, with no duplicated logic.

Ship your MCP server today.

Clone the template, deploy it, and point your agent at the URL.

View on GitHub Read the docs
  • Three transports, one codebase
  • Headless or interactive tools
  • Streamable HTTP, one port
  • Open source, self-hostable