MCP servers

MCP (Model Context Protocol) lets users extend the agent with external tools and data sources. An MCP server exposes tools over a network connection, and Eigin makes them available to the agent alongside its built-in tools.

Transport

Eigin connects over Streamable HTTP: the server exposes an HTTP endpoint, and the client sends JSON-RPC requests and receives responses or streams. It’s simple and firewall-friendly, working with cloud-hosted servers and local processes alike.

Authentication

• Open servers: no auth. Connect with a URL and go. Useful for local servers or trusted internal tools.
• OAuth 2.1 servers: the full authorization flow. Eigin discovers the server’s OAuth metadata, registers itself as a client automatically (no manual OAuth app), and sends the user to the server’s authorization page in an in-app browser to log in and grant access. Access and refresh tokens are stored securely via on-device platform APIs and refreshed automatically when they expire.

Tool discovery

On connection, Eigin calls tools/list to discover tools, each with a name, description, and parameter schema. These are cached locally so the agent sees descriptions without reconnecting every turn. Discovery re-runs on explicit refresh or when the connection is re-established.

Trust boundaries

MCP tools are external: the server is controlled by someone else, and responses could contain anything, including prompt injection. Eigin treats the boundary seriously:

• External tools default to ask: the user approves each use (see Permissions).
• Per-tool overrides let the user change the level for individual MCP tools.
• The agent sees MCP tools alongside built-in ones, but the permission system enforces the boundary.

This lets power users grant broad access to trusted servers while keeping untrusted tools behind a prompt.