• Claude
  • Cursor
  • JetBrains

Integrations · MCP editors

The same knowledge graph and control plane, from the editor you already have open.

POST /mcp over StreamableHTTP. An editor can start a change; it cannot clear its own sign-off.

What starts work

One way in, and the tools that work the change after it.

An API call starts a change. From an editor that is start_task, and it takes the same automation path a label would: the same pinned automation, the same change risk, the same set of checks.

  1. 01
    api.call

    someone calls the API for a story

  2. 02
    start_task

    starts a change on the same automation path a label would take

  3. 03
    get_task · list_gates

    reads the change and the checks it is waiting on

  4. 04
    complete_phase · log_progress

    advances a stage and writes to its log

  5. 05
    reply_comment · post_review_findings

    takes part in a review rather than only observing it

  6. 06
    file_governance · graph_query

    reads what governs a file, and queries the knowledge graph

  7. 07
    approve_gate

    refused with self_approval_refused on a change the same key started

What lands back

ADRs, patterns, gotchas and guides sit alongside the code as files, symbols, folders, communities and call edges, and file_governance returns what governs a file — the same tool an agent session receives, so your editor and the line answer “what applies here” identically.

What you grant and why

A key, shown once, authorised on every request.

There is also a stdio entry point with an older, smaller tool set. Configuring fresh, use the HTTP endpoint: the tool names differ and the line’s tools live there.

An MCP API key
Shown once when created, not retrievable afterwards, and it stops resolving the moment it is revoked.
Authorization: Bearer <key>
Each request is authorised on its own rather than on the session that opened the connection.
POST /mcp · type: "http"
StreamableHTTP. Not type: "sse", which this server does not speak.
Narrower endpoints
A tickets view, a knowledge view, a code-intelligence view — so an editor gets the tools it needs and not the rest.

An editor can start a change. It cannot clear its own sign-off.A key that started a change cannot approve its sign-off, cannot approve anything without the rights your organisation roles give it, and cannot reach an override at all; the administrative tools live on a separate endpoint.

Point your editor at the line this week.

Create a key, add one endpoint to your MCP config, and ask what governs the file you have open.