Prompty.tools

@prompty-tools/mcp

The official Model Context Protocol server for prompty.tools. Search, fetch, create, and version your prompts without leaving your AI client - Claude Desktop, Claude Code, Cursor, or any other MCP-aware host.

Quick start

Create an API key at prompty.tools/dashboard/api-keys, then register the server with your client:

# Claude Code
claude mcp add prompty -e PROMPTY_API_KEY=pk_... -- npx -y @prompty-tools/mcp

{
  "mcpServers": {
    "prompty": {
      "command": "npx",
      "args": ["-y", "@prompty-tools/mcp"],
      "env": { "PROMPTY_API_KEY": "pk_..." }
    }
  }
}

Then ask your assistant things like "find my code-review prompt on prompty and use it" or "save this as a private prompt with a direct tone".

What's in the box

33 tools, six resources

Prompts, personas, tones, outputs, constraints, and libraries - search, read, create, update, versions, and library membership.

Server-side compilation

Prompts are created from a task plus building-block IDs; the platform compiles the final text. No client-side template drift.

Safe by design

No delete tools - removal stays in the dashboard. Every create requires an explicit public/private choice, so an assistant never publishes by accident.

Typed error surfacing

Platform errors arrive with HTTP status, message, and request id. Rate-limit responses include your tier's quota so the assistant can explain instead of blindly retrying.

Quota-aware

Tool descriptions and server instructions steer assistants toward targeted searches - relevant on the free tier's 20 requests/day.

Docs that cannot drift

Every tool reference page on this site is generated from the same schemas the server registers at runtime.

Where to next

• Getting started - create a key, install in your client, verify it works.
• Concepts - the building-block model, scopes, pagination, rate limits, errors, and versioning.
• Tool reference for prompts, personas, tones, outputs, constraints, and libraries.

Looking for a TypeScript SDK instead? This server is a thin adapter over @prompty-tools/core, the typed client for the same API. Use core directly when you are writing code rather than talking to an assistant.