Automation & Ops mcpoauthrbacaccess-controlaudit-loggingself-hosting

Metorial: Identity & Access Layer for AI Agents

Open-source identity and access layer for AI agents, connecting them to 1200+ integrations with consistent auth, permissions, and observability.

FollowAgents review · FARS-2.1
Not recommended
50/ 100 5-point scale 2.5 / 5
1 2 3 4 5 6
1Trust11 / 29 · 1.9/5

Evidence: Slack integration tests explicitly verify per-tool scope gating and auth method restrictions, demonstrating least privilege (least_privilege=2). User confirmation: OAuth flow requires user authorization, but no explicit user confirmation mechanism found (user_confirmation=1). Data flow transparency: README mentions audit logs and observability, but no detailed data flow description (data_flow_transparency=1). Sensitive data handling: Tests cover token handling, but no deep discussion of encryption at rest (sensitive_data_handling=2). Dependency security: Pinned versions like bun 1.2.15, but no vulnerability scanning mentioned (dependency_security=1). External effects: Integrations call external APIs, but no side-effect control described (external_effects=1). Rollback: No rollback mechanism mentioned (rollback=0). Source attribution: README mentions Metorial company, but not verified (source_attribution=1).

2Reliability8 / 14 · 2.9/5

Evidence: Test files (e.g., Slack and Jira) show internal consistency, with clear contracts for tool keys and schemas (self_consistency=2). Dependency availability: Dependencies listed in package.json, but availability not verified (dependency_availability=1). Failure messages: Tests check error messages, e.g., Slack OAuth error (failure_messages=2).

3Adaptability12 / 18 · 3.3/5

Evidence: README targets developers and security teams, providing multiple scenarios (audience_and_scenarios=2). Capability boundaries: Tool list is explicit, but limitations not detailed (capability_boundaries=2). Trigger precision: Tool keys and schemas are test-enforced (trigger_precision=2). Environment fit: Supports multiple frameworks and languages, but deployment requirements not mentioned (environment_fit=2).

4Convention9 / 18 · 2.5/5

Evidence: README is well-structured with quick start and examples (information_architecture=2). Install notes: Provides npm and pip install commands (install_notes=2). Naming stability: Tool keys are test-enforced for stability (naming_stability=2). Examples and FAQ: Multiple examples provided, but no FAQ (examples_and_faq=2). Known limitations: Not explicitly listed (known_limitations=1). License: README claims Apache 2.0, but LICENSE file is FSL-1.1-ALv2, inconsistent (license=1). Versioning and changelog: Not mentioned (versioning_changelog=1). Maintenance responsibility: Not explicit (maintenance_responsibility=1).

5Effectiveness7 / 13 · 2.7/5

Evidence: Output usability: Tool outputs have schema definitions (output_usability=2). Marginal value: Provides 1200+ integrations, unique value (marginal_value=2). Cost-benefit: No pricing or cost mentioned (cost_benefit=1).

6Verifiability3 / 8 · 1.9/5

Evidence: Claim traceability: README claims not fully mapped to code (claim_traceability=1). Cross-source corroboration: Tests and README partially consistent, but not independently verified (cross_source_corroboration=1). Fact-inference separation: Not clearly distinguished (fact_inference_separation=1).

Evidence confidence: Low Reviewed Aug 09, 2026 Reviewed revision 8a1e799b2c38
The upstream repository has new commits since this review. The score still applies to the reviewed revision shown and may not cover the latest changes.
Safety controls not found in source: rollback or recovery path
Before you use it
  • License inconsistency: README claims Apache 2.0, but LICENSE file is FSL-1.1-ALv2, needs clarification.
  • No rollback mechanism mentioned, may affect disaster recovery.
  • Dependency security does not mention vulnerability scanning, recommend checking dependency vulnerabilities.
Review evidence [1][2][3][4][5][6][7][8]
See the full review method →

What does this agent do, and when should you use it?

Metorial is an open-source control plane that sits between AI agents and external systems, handling authentication, permissions, and observability in a unified way. It offers 1200+ integrations across SaaS tools, enterprise systems, and data sources, with built-in RBAC, SAML SSO, and IAM. Developers can expose integrations as tools to any agent framework (e.g., Claude Code, Codex, Cursor) via JavaScript/TypeScript and Python SDKs. The Metorial Platform is open-source and self-hostable, backed by Docker, PostgreSQL, Redis, and MongoDB, and a hosted SaaS is available. This repository is the integration catalog, containing MCP servers and documentation.

Metorial acts as a control plane providing unified identity and access management between agents and external systems. It indexes over 5000 MCP servers (via metorial/mcp-index), allows developers to expose integrations as tools through SDKs (e.g., @metorial/ai-sdk or metorial_pydantic_ai), handles OAuth flows via setup sessions, and applies RBAC and scoped permissions during agent sessions. It logs all actions for auditing, provides monitoring and debugging (every MCP session is recorded and inspectable), and the self-hosted version deploys locally using Docker containers for MCP servers. It does not execute agent logic itself but facilitates tool access for agents.

  1. AI application developers needing to quickly integrate SaaS tools like Slack, GitHub, or Google Calendar into Claude- or GPT-powered agents without reimplementing OAuth and permissions.
  2. Platform teams wanting to unify access credential management across multiple agents (e.g., Claude Code, Codex) while enforcing security policies.
  3. Security teams needing to audit AI agents' access to internal systems, tracking which agent acted with whose credentials for compliance.
  4. Developers wishing to test and debug MCP servers via the embedded MCP Explorer in the Metorial Dashboard, without writing code.
  5. Organizations that want to self-host the Metorial platform on their own infrastructure to maintain full control over data and access policies.
  6. Teams using multiple AI providers (OpenAI, Anthropic, Gemini) and wanting a single interface to connect the same set of tools across all of them.

What are this agent's strengths and limitations?

Pros
  • 1200+ integrations and a catalog of 5000+ MCP servers, reducing integration work.
  • Built-in RBAC, SAML SSO, and IAM simplify enterprise security compliance.
  • Audit logs track which agent used whose credentials for every action.
  • Self-hosting option gives full control over data and governance.
  • Multi-language SDKs (TypeScript, Python) support a wide range of agent frameworks like LangChain and PydanticAI.
Limitations
  • Requires learning and configuring OAuth flows, adding onboarding overhead.
  • Self-hosting needs multiple infrastructure dependencies (Docker, PostgreSQL, Redis, MongoDB).
  • As a newer project (YC F25), the community and ecosystem may still be maturing.
  • The repository is a catalog; actual functionality depends on the Metorial platform, which seems to require an API key and platform interaction.

How do you install or deploy this agent?

To use Metorial, either sign up for the hosted platform (platform.metorial.com) to get an API key, or self-host the open-source Metorial Platform from github.com/metorial/metorial-platform. Self-hosting requires Docker, PostgreSQL, Redis, and MongoDB. Install the Node.js or Python SDK: npm install metorial @metorial/ai-sdk @ai-sdk/anthropic ai or pip install metorial pydantic-ai python-dotenv. This repository itself is a catalog and does not provide direct installation steps.

How do you use this agent?

First, initialize the Metorial client: const metorial = new Metorial({ apiKey: process.env.METORIAL_API_KEY }) (Python: Metorial(api_key=os.environ["METORIAL_API_KEY"])). Create a provider deployment, e.g., providerDeployments.create({ name: 'Metorial Search', providerId: 'metorial-search' }). Then establish a session by passing an adapter (like metorialAiSdk()) and a list of providers. Use the returned session.tools() as tools for your model. For OAuth integrations, create a setup session via setupSessions.create(), show the URL to users, wait for completion, and then use the resulting providerAuthConfigId. Full examples are in the repository's examples directories.

How does this agent compare with similar options?

Compared to using MCP directly, Metorial adds a higher-layer of identity and access control, whereas MCP is just the protocol. Compared to custom-built integrations, Metorial provides ready-made connectors and a security layer.

FAQ

Is Metorial free?
The repository is open-source, and the platform offers a hosted service with a free tier, or you can self-host. For specific pricing, refer to the Metorial website.
Can I run Metorial on my own servers?
Yes, the Metorial platform is open-source and self-hostable, but it requires Docker, PostgreSQL, Redis, and MongoDB.
Which AI models does Metorial support?
Through various adapters, it supports OpenAI, Anthropic, Google Gemini, Mistral, DeepSeek, LangChain, PydanticAI, and more.
How does Metorial handle authorization and permissions?
It provides RBAC, SAML SSO, IAM, and scoped permissions per agent, workflow, team, or environment.

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related agents