openfabriq
MCP server

Tools

The nine read-only tools of openfabriq mcp, and what each one returns.

openfabriq mcp has nine read-only tools, the same as the studio's chat. Each also takes project, the repository as owner/repo, which can be left out when the machine has a single project (see Good to know).

In Claude Code, with the plugin, their names start with mcp__plugin_openfabriq_openfabriq__, for example mcp__plugin_openfabriq_openfabriq__list_runs.

ToolWhat it returnsParameters
factory_overviewThe applied factory: its agents, its automations (when they start, what they can do), its limits
list_runsThe runs, latest firstagent, status, source, subject, since, limit
read_runOne run: its record, its prompt, what the agent didid
schedulesThe scheduled automations
read_wikiThe project's wikipage, search
connectionsWhether each integration is connected
check_factory_changeWhat validate and plan would say about a changefiles
metricsThe studio overview's numbersperiod
boardWhere each subject stands

factory_overview

The factory as the engine runs it: the applied copy, not the files of your current branch. The instructions of agents and automations are left out, as they are long; the client can ask for a file when it needs one.

list_runs

ParameterValue
agentAn agent's name
statusqueued, running, succeeded, failed or skipped
sourceWhere the run came from: github, slack, linear, sentry or webhook
subjectWhat the run was about: #12, ENG-123, #factory
sinceRuns queued at or after this ISO date or time: 2026-11-01
limit1 to 200, 20 by default

read_run

The run's record (status, reason, cost, outputs proposed and their fate), the prompt the agent received, and a condensed account of what it did: its messages and tool calls. A run interrupted by a stop says which run took it up again.

schedules

For each automation with a cron: the schedule in words, the factory's time zone, the next run and the latest one with its status. A time missed while the machine slept runs once when it wakes.

read_wiki

The project's wiki. Without parameters, its folder and its pages; with page (like areas/payments.md), that page; with search, the lines that contain these words. It says so when the wiki is off or not generated yet.

connections

For Slack, Linear, Sentry, the relay and each webhook the factory uses: connected or not on this machine, and if not, the command that connects it (openfabriq connect slack). It also lists the integrations the factory does not use. It answers yes or no, never a secret.

check_factory_change

Give it each file of openfabriq/ that the change adds or modifies, with its full content (agents/reviewer/agent.md), and null for a file it deletes. It merges them with the applied factory in memory, then returns what openfabriq validate says and, when the result is valid, what openfabriq apply would change, security changes first. It writes nothing and asks GitHub nothing: the labels it announces are those the applied factory does not use yet.

metrics

The numbers of the studio's overview. period is 7d (the default), 30d or all. It returns runs by status, cost by agent and by day (the chat included), pull requests opened and merged, human interventions (handoffs, failed runs, refused outputs), the runs done without anyone stepping in, what the failed runs cost, the applied outputs by kind, and the gains GitHub tells: the factory's pull requests merged as written, and the median time from issue to merge against the team's.

board

Each subject (issue, pull request or conversation) in its stage: waiting, working, needs-you (failed, handed to a human, or stopped at a cost limit) or done, with the agents that touched it in order. Then what each agent is doing now.

On this page