ChatGPT + CircleCI: Why the Official MCP Server Doesn't Reach ChatGPT
Not out of the box. CircleCI ships an official MCP server — CircleCI-Public/mcp-server-circleci — but there’s no CircleCI-hosted remote endpoint, no mcp.circleci.com, and no app in ChatGPT’s directory. The server runs locally over stdio (NPX or Docker) or as a self-managed remote you deploy on your own infrastructure. CircleCI’s README lists Cursor, VS Code, Claude Desktop, Claude Code, Windsurf, and Amazon Q as supported clients; ChatGPT isn’t among them. So the honest answer for a ChatGPT user is: the official tooling exists, and it was built for IDEs and coding agents — not for your ChatGPT account.
Here’s what the server does, the one DIY route that gets it into ChatGPT, and where any chat-based setup runs out for CI operations.
What the official CircleCI MCP server can do
Where it does run, the server exposes 14 tools built around real CI pain:
- Get build failure logs. The flagship tool — pull the failed output from your latest pipeline and let the model diagnose it.
- Find flaky tests. Surface the tests that fail intermittently across your project’s history.
- Trigger pipelines and reruns. Kick off a pipeline or rerun a workflow from the conversation.
- Validate config. Check
.circleci/config.ymlfor errors before you push, not after the red build. - Test results and usage/cost analysis. Pull test metadata and dig into credit consumption.
Everything authenticates with a personal API token via the Circle-Token header — the same auth as CircleCI’s API v2.
The one route into ChatGPT: self-hosting
The repo’s remote mode is explicitly self-managed: you run the circleci/mcp-server-circleci Docker image on your own infrastructure with HTTP+SSE transport, set MCP_ALLOWED_HOSTS, and have each user pass their own Circle-Token header. If your team deploys that and exposes the endpoint, a ChatGPT Business or Enterprise admin can add it as a custom connector in Developer Mode.
That works — but call it what it is: a DIY deployment you operate, patch, and secure, not an official ChatGPT integration. If the transport or tool list changes upstream, you redeploy. For most teams asking “does ChatGPT connect to CircleCI,” the practical answer remains no.
Worth noting where CircleCI’s own AI energy is going instead: Chunk Sidecars, launched June 2026, give AI coding agents pre-configured cloud environments to run tests and linting before commit. CircleCI is building CI into the agent’s inner loop — not building a ChatGPT app.
The limits that actually matter
- No hosted endpoint, no directory app. Nothing to enable in ChatGPT’s settings. The official server targets IDE-class clients where local stdio is natural.
- No triggers. Even self-hosted and connected, MCP tools respond to prompts. A red build on main at 2am doesn’t wake anything — CircleCI’s
workflow-completedwebhooks exist, but ChatGPT can’t receive them. - Session-bound, even in agent mode. ChatGPT Work (launched July 9, 2026) runs long, usage-metered agent sessions — but every run is one you start manually. Useful for “dig into this failure,” useless as a standing build watch.
- Cross-stack follow-through stops at the chat. A session can explain a failure; it won’t message the on-call dev, log the incident, and schedule the follow-up on its own.
If you want CircleCI work that runs on its own: Carly
CI is the most event-driven system in your stack — builds finish on their own schedule, not yours. The valuable automation is the kind that reacts.
Carly is an AI executive assistant that acts on triggers against CircleCI’s API v2 and webhooks — no server for you to host:
workflow-completedfires with a failure on main → Carly fetches the failed job output, summarizes the root cause, and messages the on-call dev with the commit link.- Monday 8am → a flaky-test report from the Insights API, ranked by minutes wasted, in the eng lead’s inbox.
- “Is the release build green?” → Carly checks the latest workflow on the release branch and answers with status, duration, and a link.
- Nightly → trigger the E2E pipeline via the API; at 8am, a pass/fail summary with duration trend versus last week.
- No-code setup. Describe the workflow in plain English; Carly interviews you and builds it.
- Actually sends — drafts and sends email across Gmail and Outlook, posts updates, creates follow-up tasks.
- Connects to anything — 200+ native integrations, plus any other tool via your own API key.
AI agents start at $35/month, and steps in a workflow that don’t use AI run free and unlimited. Carly natively integrates with CircleCI.
ChatGPT vs Carly
| ChatGPT (self-hosted MCP) | Carly | |
|---|---|---|
| Read build logs, find flaky tests | Yes, if you deploy and maintain the server | Yes |
| Officially supported connection | No — ChatGPT isn’t a listed client | Yes, native CircleCI integration |
| Reacts to a failed build by itself | No | Yes, via webhooks and triggers |
| Weekly flaky-test digest, unprompted | No | Yes, on a schedule |
| Runs without a session open | No | Yes (cloud, 24/7) |
| Messages the on-call dev directly | No | Yes (email + chat) |
| Setup | Docker deployment + custom connector config | Describe it in plain English |
| Pricing | Paid ChatGPT plan + your hosting | AI agents from $35/mo |
ChatGPT plus a self-hosted CircleCI server is a diagnostic console you build and maintain. Carly is an assistant that watches your pipelines around the clock.
Frequently Asked Questions
Does ChatGPT work with CircleCI?
Not out of the box. CircleCI’s official MCP server (CircleCI-Public/mcp-server-circleci) runs locally over stdio or as a self-managed remote you deploy yourself — there’s no CircleCI-hosted endpoint and no ChatGPT directory app, and ChatGPT isn’t in the repo’s supported-client list. A team can self-host the remote mode and add it to ChatGPT as a custom connector, but that’s a DIY deployment.
What can the official CircleCI MCP server do?
It exposes 14 tools: fetching build failure logs, finding flaky tests, triggering pipelines and reruns, validating config files, pulling test results, and usage/cost analysis. It authenticates with a personal API token and officially supports clients like Cursor, VS Code, Claude Desktop, Claude Code, Windsurf, and Amazon Q.
Can ChatGPT alert me when a CircleCI build fails?
No. CircleCI offers workflow-completed and job-completed webhooks, but ChatGPT can’t receive webhooks — it only acts inside sessions you start. For failure alerts with a summarized log, you need a trigger-based assistant like Carly.
Does CircleCI have its own AI features?
Yes, aimed at coding agents rather than chat: Chunk Sidecars (June 2026) give AI agents pre-configured cloud environments to run tests and lint before committing — CI validation inside the agent’s inner loop.
More: Claude + CircleCI · ChatGPT + GitHub · ChatGPT + GitLab · ChatGPT + Docker Hub · ChatGPT + Sentry · Best AI workflow automation tools
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
