MCP tools overview
Model Context Protocol integrations—connect external tools and data to your agents.
MCP tools overview
What is MCP?
MCP (Model Context Protocol) integrations let agents use external tools and resources (Slack, GitHub, Google Drive, Notion, Confluence, etc.) through a unified tool service and transport layer in the backend.
Dashboard
Dashboard → Connectors (/mcp-integrations) lists available connections and OAuth completion routes for providers that need it.
OAuth completion routes
The app includes completion pages after OAuth, for example:
/mcp-integrations/slack/complete/mcp-integrations/github/complete/mcp-integrations/google/complete/mcp-integrations/notion/complete/mcp-integrations/confluence/complete
Finish the browser flow in one session; don’t share completion URLs.
Sessions & registry
The backend maintains integration registry, sessions, and tool execution paths so the agent runtime can call MCP tools safely alongside built-in tools.
Security
Secrets and tokens are stored with appropriate encryption and access controls—only admins should connect or revoke integrations.
See individual guides: Slack, GitHub, Google, Notion, Confluence.
For how isolated execution fits agent and MCP architectures end-to-end, read E2B sandboxes for AI agents on the blog (it links back to Tool calling & agentic execution and Workflows).