The check your coding agent can call, before the pull request opens.
Skene runs as a hosted MCP server: eleven tools your agent calls in the same session it is already in, scoped by the API key you issue.
MCP is the protocol your agent uses to call outside tools. The hosted endpoint is www.skene.ai/api/mcp; auth is a workspace API key.
Agent
Complete. The feature works and the tests pass.
Check
signup_completed is no longer recorded on this path.
Runs in the same session
Not a separate tool to open or a second dashboard to check. One more thing the agent can call mid-turn.
The gate is deterministic
Returns the events your journey says should exist and do not. Call it over HTTP or as an MCP tool.
Scoped by the API key you issue
Hosted by Skene. The key belongs to the workspace, and you issue it.
A check the agent did not write.
endpoint www.skene.ai/api/mcp auth workspace API key
- An event the journey plans for that no longer fires
- A database event whose table is not there
- A tracking call for an event nobody planned
- A planned property the call stopped sending
Missing event- fails gate
Missing table- fails gate
Unplanned tracking call- reported
Dropped property- reported
An agent checking its own tracking is checking its own work.
The MCP check compares your diff against a baseline the agent never touches: the planned event baseline from your workspace journey artifacts.
Two paths that can disagree.
The MCP check and the pull request review are separate: one is a deterministic comparison against your journey baseline, the other a model reading the diff against a written policy. They share no code.
- Either can be used alone; neither replaces your test suite
- Both are about tracking specifically, which most CI suites do not check at all
For coding agents: Cursor, Claude Code and any other MCP-capable agent. Add the hosted endpoint to the agent’s config; the check is callable the same turn.
Put the check in your agent's reach.
No card needed, minutes to connect.