Mergify MCP server: use Agent Skills for merge queues
Mergify runs merge queues and merge protections for GitHub pull requests. Agents reach it through skills that run the Mergify CLI, not through MCP.
Checked 3 October 2026 · from Mergify's Agent Skills documentation, its llms.txt, its OpenAPI document and the Mergifyio GitHub organization
One of 91 guides to monitoring and developer tools. Each covers what an AI coding agent can get out of the tool through MCP, the protocol that lets the agent query the tool directly.
Summary
Mergify has not published an MCP server as of 3 October 2026. Its machine-readable documentation index, llms.txt, has no MCP entry, and the official MCP registry returns nothing for Mergify.
Agents reach Mergify through Agent Skills instead. A skill is a bundle of instructions that teaches a coding agent to use a tool, here the mergify command-line interface (CLI).
The skills answer the questions a stuck merge raises: queue position, the reason for a dequeue, and what a rule would do. Mergify documented them on 19 August 2026.
How Mergify works with MCP
Mergify automates merging on GitHub. Its merge queue tests pull requests against the latest base branch, often in batches, before they land.
Merge protections add freezes, pull request dependencies and scheduled merges. CI Insights collects test results and flags flaky tests.
The skills live under skills/ in Mergifyio/mergify-cli and are Apache-2.0 licensed. Skill files first appear in that repository's history on 10 April 2026.
Six skills ship as one bundle: mergify-merge-queue, mergify-events, mergify-config, mergify-merge-protections, mergify-ci and mergify-stack.
Each skill shells out to the CLI, and the CLI calls Mergify's REST API. That API is described by an OpenAPI 3.1 document and includes GET /repos/{owner}/{repository}/merge-queue/status.
Every vendor here gets the same four checks: official, maintained, distributed and usable. Mergify passes 0 of them.
| Check | Result | Evidence |
|---|---|---|
| Official | Fail | On 3 October 2026 Mergify's documented agent route is a bundle of Agent Skills that run its CLI. It has not published an MCP server. |
| Maintained | Not applicable | Not applicable without a server. The skills ship in Mergifyio/mergify-cli, last pushed on 2 October 2026, with release 2026.10.1 published the same day. |
| Distributed | Not applicable | The official MCP registry returned no Mergify entry on 3 October 2026. The skills install from the Claude Code plugin marketplace, the Codex plugin directory or skills.sh. |
| Usable | Not applicable | Not applicable to MCP. The skills need a terminal agent that can run the mergify binary after mergify auth login. |
| Part of the work | What Mergify has here | Can an agent reach it |
|---|---|---|
| Code, tests & review | Rules in .mergify.yml, stacked pull requests, CI Insights test results | Yes, through skills. mergify-config validates the file and simulates rules on a real pull request |
| CI & release | Merge queue, batches, dequeue reasons, freezes, Activity Log | Yes, through skills or the REST API. Queue status, dequeue causes, pause and resume |
An agent can explain why a pull request has not merged, as long as it can run a shell. In a chat client without one, the skills do nothing and the REST API is the remaining route.
What can an AI agent do with Mergify?
Every Mergify skill fails on its first command
The skills are instructions for the mergify binary, not an API client. If the binary is missing from the agent's PATH, or nobody has signed in, the first call fails.
Guard: Install the CLI and run mergify auth login before installing the skills. On macOS, install with brew install mergifyio/tap/mergify-cli.
A command fails with an application key that works for other commands
Mergify application keys carry either the admin scope or the ci scope. admin covers queue status, the Activity Log and freezes.
CI and test uploads, Merge Queue Scopes and mid-run test-selection reads need a ci key instead. Neither scope is read-only.
Guard: Choose the scope for the job the agent will do. A key reaches every repository its GitHub account can see, so prefer mergify auth login on a workstation.
Stack pushes fail after exporting an application key
Stacks commands look for --token, then MERGIFY_TOKEN, then GITHUB_TOKEN. An application key in MERGIFY_TOKEN is handed to GitHub, which rejects it.
Guard: Pass a GitHub token with --token on Stacks commands, or keep the application key out of MERGIFY_TOKEN.
Do you need the MCP server at all?
Without a shell, call the REST API at https://api.mergify.com/v1 with a bearer token. An agent with an HTTP tool can read queue status or one pull request's queue entry.
For the pull request around the queue, GitHub's MCP server reads checks, comments and merge commits. Use it alongside the skills when a question spans both systems.
Mergify does use MCP internally. A 20 February 2026 post describes its support triage built on Claude Code, Sentry, Linear and local MCP servers, all consumed rather than offered.
Use the skills in a terminal agent. Use the REST API from anything else.
Mergify integration setup
The bundle is the same in every client. Install and authenticate the CLI first, because each skill depends on it.
Authenticate the CLI
mergify auth login opens an approval page and stores a Mergify user token outside the agent's reach. Add --no-browser on a machine without a browser.
brew install mergifyio/tap/mergify-cli mergify auth login
Claude Code
The claude-plugins-official marketplace ships with Claude Code, so the Mergify plugin installs with one command.
/plugin install mergify@claude-plugins-official
Cursor, Cline, Copilot and other skills.sh agents
skills.sh detects the agent it runs under and writes the skills where that agent reads them. Add --agent cursor --global to pick the target and cover every project.
npx skills add Mergifyio/mergify-cli
What the vendor documents
Mergify's documentation suggests plain requests, such as asking why pull request #1234 left the merge queue, or freezing the queue until Monday morning. The agent picks the matching skill and runs the CLI.
Should you connect an agent to Mergify?
Yes, through Mergify's Agent Skills, if your agent can run a shell. Best for teams on Claude Code, Codex or Cursor who want merge queue answers in the terminal.
- Strongest fit
- Yes, through Mergify's Agent Skills, if your agent can run a shell.
- Main advantage
- Queue status and dequeue reasons come from Mergify itself, not from side effects on the pull request.
- Main weakness
- There is no MCP endpoint, so browser-only chat clients cannot use the skills.
- Operational risk
- Application keys are not read-only and reach every repository their GitHub account can see.
From Oodle
One platform for agent traces and infrastructure
Agent traces usually sit in a different product from the rest of your telemetry, so when a slow database makes an agent slow the symptom is in one tool and the cause is in another. Oodle keeps both in one query surface.
See agent observabilityFrequently asked questions
Does Mergify have an MCP server?
No. As of 3 October 2026 Mergify has not published one. It ships Agent Skills that run its CLI, and it documents a REST API.
Can an agent find out why a pull request left the merge queue?
Yes. The mergify-merge-queue skill works out why a pull request was dequeued and which check failed. mergify-events reconstructs its full queue lifecycle from the Activity Log.
Do the Mergify skills work in ChatGPT or Claude on the web?
No. They shell out to the mergify binary, which a browser chat cannot run. Mergify's Codex listing works only from Codex for the same reason.
Is there an API an agent can call directly?
Yes. The REST API base URL is https://api.mergify.com/v1, and its OpenAPI 3.1 description is at docs.mergify.com/openapi.json.
