How to Connect CircleCI to Claude (and What It Can't Do)
CircleCI was one of the first CI vendors to publish its own MCP server — an official, open-source one that CircleCI itself maintains. So the short answer to “does Claude work with CircleCI?” is yes, and you’re not depending on a stranger’s side project to make it happen. The longer answer has two catches: CircleCI isn’t a one-click app in Claude’s connector directory as of mid-2026, and once connected, Claude can only look at your builds when you’re in a chat asking about them. It will never notice a red pipeline on its own.
What CircleCI’s official server gives Claude
CircleCI built its MCP server around the question developers actually bring to a chat: why is my build broken? Per CircleCI’s docs, the server hands Claude structured access to build logs, job metadata, and failure context, which unlocks things like:
- Diagnosing a failing build from the actual log output, not a guess
- Spotting flaky tests — the ones that fail randomly and erode everyone’s trust in CI — from your project’s test history
- Summarizing what a pipeline did across its workflows and jobs
If you’re a founder who hears “the build is red again” in standup and nods along, this is the appeal: paste nothing, just ask.
“Why did the latest workflow on the checkout-service project fail?”
“Which tests in this project have been flaky over the past month?”
“Compare the last green build on main with the current failing one — what changed?”
Setup: easier in Claude Code than in claude.ai
Here’s a wrinkle specific to CircleCI. The official server is distributed as a package you run locally (npx @circleci/mcp-server-circleci with a CIRCLECI_TOKEN environment variable). That’s a natural fit for Claude Code and Claude Desktop, which happily run local MCP servers.
Getting it into claude.ai on the web is clunkier: web custom connectors expect a remote server address, so you’d need to host the CircleCI server somewhere reachable first. Whichever surface you use, custom connectors are gated to paid Claude plans. The steps, roughly:
- Generate a personal API token in your CircleCI user settings.
- Local route: add the server to Claude Code or Claude Desktop’s MCP config with your token.
- Web route: deploy the server to a host you control, then add its URL under Settings → Connectors → Add custom connector.
- Ask about a pipeline to confirm it’s wired up.
Ten minutes if you live in a terminal. A real project if you don’t — which matters, because the people who most want plain-English answers about CI are often the least terminal-inclined people on the team.
A red build at 2 a.m. doesn’t start a conversation
Every MCP connector shares one boundary, and with CI it stings more than usual: builds fail at inconvenient times, and Claude only works while you’re actively chatting with it. Concretely:
- A deploy workflow fails at 2 a.m. Claude finds out when you do — the next morning, when you open a chat and ask.
- A flaky test starts failing every third run on Tuesday. Nobody asks Claude about flaky tests until Friday, so it stays invisible for three days.
- You want release notes posted after every successful deploy pipeline. Claude can write them beautifully — but only if a human requests them, every single time.
There’s no mechanism for “when a job fails, do X.” The connector is a diagnostic partner for the build you’re staring at, not a monitor for the ones you aren’t.
Closing the loop with Carly
That “when a job fails, do X” sentence is precisely what Carly exists for. Carly is an AI executive assistant that runs trigger-based workflows in the cloud, around the clock. Wired to CircleCI, a few workflows teams actually build:
- Failed pipeline on main → Carly summarizes the failure in plain English, emails the commit author, and files a Linear or Jira ticket — sent and filed for real, not drafted.
- Deploy pipeline succeeds → release notes go out to the team channel without anyone asking.
- Friday afternoon → a digest of the week’s build health (including which tests flaked) lands in your inbox.
You don’t configure any of this with YAML or tokens-in-terminals. You tell Carly, in plain English, “when a CircleCI pipeline fails on main, email the author a summary and open a ticket,” and it asks clarifying questions, then builds the workflow with you. AI agents start at $35/month; steps that don’t use AI run free and unlimited. CircleCI sits alongside 200+ other tools — see integrations.
Which tool for which job
| Claude (CircleCI MCP) | Carly | |
|---|---|---|
| Diagnose the failing build you’re looking at | Yes — its sweet spot | Yes |
| Surface flaky tests when asked | Yes | Yes |
| Act the moment a pipeline fails | No — someone must open a chat | Yes, trigger-based |
| Post release notes after every green deploy | Only if prompted each time | Yes, automatically |
| Works from claude.ai without hosting a server | No (local server; web needs hosting) | n/a — nothing to host |
| Pricing | Paid Claude plan | AI agents from $35/mo |
Frequently Asked Questions
Does Claude integrate with CircleCI?
Yes. CircleCI publishes an official open-source MCP server that gives Claude access to build logs, job metadata, and test history. It’s not a one-click app in Claude’s connector directory as of mid-2026 — you run the server yourself (easiest in Claude Code or Claude Desktop) and connect it, which on claude.ai means a custom connector and a paid plan.
What’s the best thing to use it for?
Build diagnosis. Asking Claude why a workflow failed, or which tests are flaky, gets you a plain-English answer sourced from real logs — genuinely faster than digging through the CircleCI UI, especially for non-engineers.
Can Claude notify me when a CircleCI build fails?
No. MCP connectors have no event triggers — Claude only acts inside a conversation you start, so a failed job at 2 a.m. goes unnoticed until someone asks. For automatic reactions to pipeline events, use a trigger-based agent like Carly.
Do I need a paid Claude plan for this?
For claude.ai, yes — custom connectors are a paid-plan feature. Claude Code and Claude Desktop can run the server locally, but those sit on paid plans too. Budget for a Claude subscription either way.
What if I want the whole loop — detect, summarize, notify, ticket — without touching it?
That’s a Carly workflow, not a Claude connector. Carly fires when the pipeline event happens, runs each step in the cloud, and actually sends the email and files the ticket. AI agents start at $35/month.
More: Claude connectors · Claude vs Carly · Can Claude send emails · Claude models explained · Claude + Cloudflare · Claude + Datadog · Claude + Vercel
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."
