The open, community-driven registry for Model Context Protocol (MCP) tool servers.

MCP Foundry catalogs, scans, and distributes MCP tools: the building blocks that let AI agents interact with the real world. Whether it's querying a database, managing GitHub issues, or automating a browser, every tool in the registry is transparent about its permissions and reviewed for safety.

The Model Context Protocol is an open standard created by Anthropic that gives AI assistants a universal way to call external tools. An MCP server exposes functions (called "tools") over JSON-RPC via stdio, and any compatible AI agent can discover and invoke them.

MCP Foundry exists to solve the discovery and trust problem: How do you find tools? How do you know they're safe?

Our registry is the canonical source of truth for MCP tool metadata. It currently tracks 62+ tools across 8 categories:

23 verified tools have passed automated security scanning and maintainer review. The remaining community tools are cataloged with full permission transparency while awaiting verification.

┌─────────────────┐     ┌──────────────────┐     ┌─────────────────┐
│  Developer       │     │  MCP Foundry      │     │  AI Agent        │
│  submits tool    │────▶│  Registry          │────▶│  discovers &     │
│  via PR          │     │  scans & reviews   │     │  installs tools  │
└─────────────────┘     └──────────────────┘     └─────────────────┘

1. Submit: Fork the registry, add your toolbox.toml manifest to submissions/, open a PR
2. Scan: Our CI pipeline validates the manifest structure and runs security analysis
3. Review: Maintainers verify permissions, check for prompt injection patterns, and approve
4. Publish: The tool appears in the verified catalog, available to every agent and runtime

The catalog is served as static JSON at registry.mcpfoundry.org:

GET https://registry.mcpfoundry.org/verified.json     # Security-verified tools
GET https://registry.mcpfoundry.org/unverified.json   # Community tools (awaiting review)

Both endpoints support CORS, return application/json, and are cached at the edge with 5-minute TTL.

Every tool in the registry declares its permissions explicitly:

• Filesystem: Which paths it needs to read/write (or false for none)
• Network: Which domains/protocols it connects to (or false for none)
• Processes: Whether it spawns child processes

Verified tools are additionally scanned for:

• Prompt injection patterns in tool descriptions
• Overly broad permission requests
• Known malicious code signatures
• Mismatched declared vs. actual behavior

Tools from the registry work with any MCP-compatible client, including Claude Desktop, Cursor, Agent Toolbox, Windsurf, and others.

We welcome contributions of all kinds:

• Submit a tool: Open a PR with your toolbox.toml manifest
• Report issues: File a bug or security concern
• Improve the registry: Better scanning, more categories, documentation fixes

See the registry README for detailed submission guidelines.

• Website: mcpfoundry.org
• Registry API: registry.mcpfoundry.org
• MCP Spec: modelcontextprotocol.io