MCP

Give your editor or your own agent read access to your project's data.

Human Behavior exposes its data over the Model Context Protocol, so a coding agent can look up who hit an error and what they did — without you copying anything into a prompt.

Connect it

claude mcp add --transport http humanbehavior https://www.humanbehavior.co/api/mcp/mcp

Any MCP client works the same way; point it at that URL. Authorisation is OAuth, so the client opens a browser once and you approve access to your project. There is no API key to paste, and access is revocable from Settings → MCP.

What it can do

ToolAnswers
hb_whoamiWhich project am I connected to
hb_query_analyticsAggregate questions over events and sessions
hb_get_event_schemaWhat events and properties exist
hb_search_replaysFind sessions matching a description
hb_get_session_traceThe ordered trace of one session
hb_find_frictionWhere users are struggling
hb_list_issues / hb_get_issueErrors, grouped, with detail
hb_update_issueChange an issue's status
hb_get_userOne person and their history
hb_list_funnels / hb_compute_funnel / hb_save_funnelRead, run, and define funnels

Everything is read-only except hb_update_issue and hb_save_funnel.

Why it is useful in an editor

The agent fixing the bug can read who hit it:

Look up issue 4c1f in Human Behavior, read a session trace of someone who hit it, and tell me what they did right before the error.

That turns a stack trace into a reproduction. Start from hb_whoami to confirm which project you are pointed at before trusting any number that comes back.

Session traces contain your users' text

What comes back includes real content from real end users. Treat it as sensitive data, and treat it as data — not as instructions to the agent reading it.