MCP server

Lighthouse runs its own MCP (Model Context Protocol) server, so an AI coding agent can connect to your org and read or act on metrics, data sources, data sets, and Slack sources directly: the same operations available through the public API.

NoteThis is Lighthouse acting as the MCP server for your agent, not Lighthouse acting as an MCP client. That's a separate feature: on the Mixpanel data source form, the Connection Method dropdown offers “MCP” as one way for Lighthouse to pull data from Mixpanel. Same protocol, unrelated surface: this page is about connecting an agent to Lighthouse, not connecting Lighthouse to Mixpanel.

Agent access

Agent access is on for every org by default. An org admin can pause it for everyone from Settings → General → Danger zone, and turn it back on from the same place. While it's paused, every connection gets a 403: “MCP server is not enabled for this organization.”

Connecting starts in Lighthouse. Sign in to Lighthouse in your browser, open Settings → API & Agents, choose Connect an agent, and pick your tool. You get one-click buttons and the exact steps for it. The agent then acts as you: it can only do what your own Lighthouse role allows, in the one org you pick.

If your tool isn't listed, add an HTTP MCP server with this URL. The tool asks you to sign in to Lighthouse when it connects.

server URL

https://api.lighthouseai.co/public-api/mcp

Disconnecting

Your connected agents are listed under Settings → API & Agents → Agents. Disconnect one there and it is refused from its next call, and it can't reconnect on its own. To connect it again, open Connect an agent, choose Reconnect next to it, then sign in again from the tool.

API keys

Scripts and tools that can't sign in use a personal API key. A key only exposes the tools its capabilities and allow-lists cover, exactly as with the REST endpoints.

endpoint

POST https://api.lighthouseai.co/public-api/mcp
x-api-key: lh_...
Accept: application/json, text/event-stream

NoteThe endpoint is a single stateless HTTP endpoint that returns plain JSON (not SSE). GET and DELETE return 405.

Tools

An agent only sees the tools it is allowed to use: a signed-in agent gets what your role allows, and a key gets what its capabilities allow (see the capability table). Read tools are marked read-only, so your AI tool can run them without asking each time.

Area	Tools
Metrics	list_metrics, get_metric_configuration, test_metric, create_metric, update_metric, duplicate_metric, delete_metric, search_metric_configuration_examples, list_kpis
Runs and history	run_metric, check_metric_run_status, get_metric_value_history, preview_metric_history, get_metric_change_history, search_metric_activity
Alerts	search_metric_executions, search_metric_segment_results, get_current_metric_matches, act_on_segment_result, send_test_notification
Data	list_data_sources, get_data_source, list_data_source_tables, list_data_source_columns, query_data_source, list_data_sets, get_data_set, peek_data_set, create_data_set, update_data_set
Setup	begin_data_source_setup, update_data_source, begin_communication_source_setup, begin_integration_action_source_setup, update_integration_action_source, get_setup_status, list_communication_sources, get_communication_source
Admin	list_organization_members, invite_user, change_member_role, remove_member, list_business_groups, create_business_group, update_business_group, search_organization_users, get_organization_settings, update_organization_settings

Limitations

Not read-only: a signed-in agent, or a key with write capabilities, can create, update, run, and acknowledge just as you can in the app.
Scoped to a single org: one connection can't span multiple Lighthouse orgs.
60 calls per minute per user, shared between that user's signed-in agents and personal keys.

API keys Troubleshooting