SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI agentsmetorial

metorial

Metorial (YC F25) is an open-source identity and access layer tool for AI agents. It acts as a control plane between agents and external systems, handling authentication, permissions, and observability in a unified way. It supports connections to over 1,200 integrations (covering SaaS tools, enterprise systems, data sources, and custom MCP servers), provides lifecycle management for credentials such as OAuth, API keys, and service accounts, and includes built-in RBAC, SAML SSO, IAM, and fine-grained permissions scoped by agent, workflow, team, or environment. Metorial can log and audit agent actions, tracing which agent performed an action using whose credentials. For developers, it provides a unified API and connection URL that can seamlessly integrate with Claude Code, Codex, Cursor, and various agent frameworks (such as Vercel AI SDK, OpenAI, Anthropic, LangChain, LangGraph, LlamaIndex, AutoGen, CrewAI, etc.). The platform itself is open source and self-hostable, while also offering a hosted platform option.

Open website

About this tool

Metorial is an open-source identity and access layer designed for AI agents to connect to external systems. It is positioned as a control plane for agent access to external systems, sitting between agents and various integrations to handle authentication, permissions, and observability in a consistent way, avoiding the need for teams to write ad hoc integration and authentication logic for each application.

In terms of purpose, Metorial solves the problem of the lack of unified identity and access management when agents connect to production systems: limited access control, missing auditing, and no standard access restrictions. It enables agents to connect to real systems under controlled, auditable, and traceable conditions.

Key capabilities include:

  • Integration connections: supports 1,200+ integrations, covering SaaS tools, enterprise systems, data sources, and custom MCP servers.
  • Authentication and token lifecycle management: supports OAuth, API keys, service accounts, and other credential flows.
  • Permission and security model: built-in RBAC, SAML SSO, and IAM, with support for scoped permission control by agent, workflow, team, or environment.
  • Auditing and traceability: logs agent action records, including which agent performed an action using whose credentials.
  • Shared access patterns: reuse connections across teams and projects instead of writing one-off integration code for each.
  • Developer experience: provides a unified API and connection URL (magic URL), unifying authentication flows such as OAuth and API keys; integrations can be exposed as tools to any agent framework.
  • Security and compliance: centralized access control, clear permission boundaries, controlled integration sharing, and consistent credential handling make agent activity enforceable and inspectable.
  • MCP support: built on Model Context Protocol (MCP), provides an index of 5,000+ MCP servers, supports testing and exploring MCP servers in the embedded MCP Explorer within the Dashboard, and supports monitoring and debugging MCP sessions, automatically detecting errors and providing detailed reports.

In terms of usage, Metorial offers both a hosted platform (you can sign up for free to get started) and a self-hostable open-source platform. The platform is powered by the MCP server in this repository. Developers can use the JavaScript/TypeScript or Python SDK to integrate through the unified API, and can refer to the official examples directory (including typescript-quick-start, typescript-ai-sdk, typescript-openai, typescript-anthropic, typescript-google, typescript-openai-compatible, typescript-langchain, typescript-provider-config, pydantic-ai, langchain, langgraph, openai-agents, llamaindex, autogen, crewai, google-adk, haystack, etc.). For services that require user authentication (such as Slack, GitHub, Google Calendar, or SAP), Metorial provides setup session to handle the OAuth flow: create a session via providerDeployments.setupSessions.create() (TypeScript) or provider_deployments.setup_sessions.create() (Python), display the URL to the user to complete browser authentication, wait for the session to complete and return the auth config, then pass its ID into the provider configuration.

In terms of the tech stack, Metorial's core runs on Bun, most of the code is TypeScript, the MCP engine is written in Go, data storage uses PostgreSQL, caching and real-time data processing use Redis, usage and logs are stored in MongoDB, the Dashboard is built with React, and MCP servers run via Docker. Repository components include: Metorial (integration catalog), Metorial Platform (core engine, open source and self-hostable), Metorial CLI (CLI for agents), Lowerdeck (shared library), Starbase (MCP debugging and testing tool). The Catalog uses the Apache License 2.0.

Applicable scenarios include: developers who want their AI applications to connect to any API, data source, or tool in a reliable, simple, and secure way; security teams that need centralized control, auditing, and permission management over agent access to systemsboundary delineation.

Limitations and information not provided: the introduction does not state specific pricing, details of free quotas, whether all integrations are available or require additional authorization, or details such as deployment requirements beyond the hosted platform; these have not yet been provided.

Related tools

Browse category