corsairdev/corsair is a library that integrates AI agents with external applications through built-in permission layers, scoped authentication, and webhook handling without exposing credentials. Agents never see API keys; credentials are resolved internally at call time.
Project overview
It intercepts agent tool calls to create database-backed approval requests that the agent cannot bypass, storing permission requests in a database the agent cannot access.
Project type
MCP · AI Agent · Infrastructure
Use cases
Automation
Deployment
Refer to project documentation
License
License pending
Best for
Developers integrating AI agents with external applications who need to restrict agent actions through configurable permission modes and credential isolation.
Enterprise teams that need per-tenant isolation of credentials, data storage, and permissions.
Key capabilities
Connect the Corsair MCP to any agent to provide built-in tool calls, permissions, and scoped auth.
Configurable modes (open, cautious, strict, readonly) for integrations to control or block agent actions.
Provides isolated credentials, data storage, and permissions per tenant when enabled.
Ships typed, signature-verified webhook handlers routed to a single endpoint.
Stores credentials in an encrypted database using envelope encryption with a user-controlled Key Encryption Key.
Usable as a library where permission layer and key management are needed, dropping down to individual SDKs when necessary.
Intercepts agent tool calls to create database-backed approval requests that the agent cannot bypass.
Limitations and risks
New integrations are being added regularly; the full list is maintained in the project documentation rather than enumerated exhaustively here.
Getting started
Install the corsairdev/corsair library, configure plugins and multi-tenancy settings, then integrate the agent. Setup is rated medium difficulty because it requires implementing library code and configuring external plugins.
Evidence and sources
README: Corsair is the unified integration layer for your agents.
README: Corsair is a library. Use it where the permission layer and key management help, and drop down to individual SDKs when you need custom logic.
README: Corsair allows you to safely integrate with any app. Connect the Corsair MCP to any agent and have built-in tool calls, permissions, and scoped auth.
README: Each integration has its own mode. Set GitHub to strict and Slack to cautious based on how much you trust each surface.
README: Set `multiTenancy: true` and every tenant gets isolated credentials, isolated data storage, and isolated permissions handling. You can scope a request to a tenant id and Corsair e…