MCP setup

Connect your agent to Core Edge.

Core Edge is CLI-first, but agents that cannot use a shell can still access a Core namespace through hosted or local MCP.

Endpoint

https://core-edge.up.railway.app/mcp

Choose a connection path

Client capabilityUse this pathAuthentication
Can run terminal commands Use the core-edge CLI. After installation, core-edge skill --install can install the matching local skill. core-edge login with browser approval for humans. Agents use CORE_AUTH_TOKEN.
Can call hosted Streamable HTTP MCP with bearer auth Point the client at the hosted /mcp endpoint. Configure Authorization: Bearer <core-key> or the bearer-token field your client provides.
Supports OAuth/Mixed Auth for remote MCP Use OAuth with Dynamic Client Registration when offered. Authorize with a scoped machine key in the Core Edge authorization page.

Learn the operating model

Connection is only the first step. Read the complete agent operating guide for spaces, search-before-write, typed records, links, the first useful shared context, and second-agent continuity.

Create a Core key

An admin creates an ordinary Core Edge key for the agent. The mcp purpose label is provenance only; role and namespace membership are the access boundary.

core-edge keys create my-agent --role write --purpose mcp

Use read instead of write for read-only connectors.

OAuth / DCR clients

For remote hosted clients that support OAuth or Mixed Auth, use these values:

Discovery endpoints:

If your client shows a generated ce_dcr_... client ID, do not paste that into the Core key field. Use a scoped machine key created for that client.

Security boundary

MCP receives the same namespace-scoped access as the Core key. Key management, hard delete, approval review, migrations, raw arbitrary API requests, and operator controls are not exposed through the default MCP tool surface.