MCP Connectors
An MCP connector gives one Hatcher agent access to a trusted remote Model Context Protocol server. Connector configuration is owner-managed and tool access is explicit.
Add a connector
Open the agent
Go to My Agents, open an agent, then select the Connectors tab.
Enter the endpoint
Add the connector name and its trusted HTTPS MCP endpoint. Use the authentication option required by that server.
Discover tools
Save the connector so Hatcher can read its tool catalog. A successful discovery shows the number of available tools.
Set the allowlist
Enable only the tools the agent needs. Start with read-only tools, then add effectful tools deliberately.
Test through a bounded mission
Run a small task and inspect the request in Mission Control. Effectful calls appear in Trusted Actions unless covered by a matching grant.
Security behavior
- Connectors are scoped to one owned agent.
- Authentication material is stored encrypted and is not returned to the browser after save.
- Revoking a connector removes its future tool access.
- Tool discovery does not authorize every discovered tool.
- Effectful execution is separately gated by approval policy.
Connect only MCP servers you trust. Tool descriptions and results come from the remote server, and an approved effectful tool may change external state.
Troubleshooting
If discovery fails, verify the HTTPS URL, authentication, server availability, and that the endpoint returns a valid MCP tool catalog. If a tool is visible but cannot run, check the connector allowlist and the Approvals inbox.