Metorial Wants to Replace Every Agent's OAuth Mess With 1200+ Integrations
Metorial is an open-source control plane that gives AI agents standardized auth, permissions, and audit logs across 1,200+ integrations via MCP.
Takeaways
- Metorial is an open-source identity and access layer for AI agents, powered by MCP, connecting them to 1,200+ integrations.
- Handles OAuth, API keys, RBAC, SAML SSO, and audit logs so agents inherit your existing security model.
- SDKs for TypeScript and Python with adapters for OpenAI, Anthropic, Gemini, LangChain, PydanticAI, CrewAI, and more.
- MCP servers run in isolated Docker containers, with an embedded MCP Explorer for testing before writing code.
- Apache 2.0 licensed, self-hostable via Metorial Platform, or use the free hosted tier.
- Backed by Y Combinator, now over 3.3k GitHub stars and 367 forks.
Wiring an AI agent up to Slack, GitHub, Stripe, and half a dozen internal APIs usually means writing bespoke OAuth flows, hardcoding tokens, and hoping nobody audits the mess later. Metorial, a Y Combinator-backed project that just crossed 3.3k stars on GitHub, wants to collapse that entire layer into a single control plane that any agent framework can plug into.
The pitch is straightforward: an open-source identity and access layer for AI agents that connects them to real systems with consistent auth, permissions, and observability. Under the hood it runs on the Model Context Protocol and ships Docker-containerized MCP servers so each integration runs isolated.
The problem it actually solves
Agents are being connected to production systems without a consistent identity and access layer around them: limited access control, no auditing, no standard access restrictions. If you have shipped anything agentic to production, you have probably felt this. Every new integration is a fresh OAuth dance, a fresh secret to rotate, and a fresh blind spot for security review.
Metorial sits between agents and integrations as a shared layer. Instead of every team reimplementing token refresh logic for the same twenty SaaS APIs, you get one interface that handles it and logs which agent did what with whose credentials.
What ships in the box
- 1200+ integrations across SaaS tools, enterprise systems, data sources, and custom MCP servers