mrkeyoor.com_
Tue 01 Sept 16:16 UTC
Dev ToolsMCP servermeasured 19 Aug 2026

CircleCI

CircleCI's own MCP server, published as @circleci/mcp-server-circleci, gives an assistant stdio access to your CircleCI account: pipeline status, build failure logs, flaky test detection, artifacts, reruns, rollbacks and usage data. It boots and lists 13 tools. It also prints a deprecation banner on startup, pointing at CircleCI's hosted MCP server instead.

BootBOOTS ✓We started it and initialize came back in 17.5s.
Tools13tools/list payload 51,736 bytes
Context tax~12,932estimated tokens for the whole tools/list, chars/4 estimate
Resources / prompts / counts returned by resources/list and prompts/list
Measured19 Aug 2026one run of the harness, clean environment, no credentials · how we test

How we started it

npx -y @circleci/mcp-server-circleci
Verdict

Use it only if you are pinned to a self-hosted stdio setup and you accept that the package tells you, in a box drawn with line characters, that it will stop getting security fixes. The tools themselves are the right ones for CI work: get_build_failure_logs and find_flaky_tests are exactly what you want an agent reaching for at 2am. But 13 tools cost more of your context window than most whole servers in this batch, and the vendor is asking you to move.

Use it if

You debug red builds through an assistant and want it to pull failure logs and test results itself instead of you pasting them
Flaky tests are a standing argument on your team and you want find_flaky_tests queried on demand rather than in a weekly report
You need an agent that can rerun a workflow or trigger a rollback pipeline without you opening the CircleCI UI
Your client only speaks stdio and the hosted server's transport is not an option for you yet

Skip it if

You read the startup banner and, reasonably, do not want an MCP server that has announced it will stop receiving security fixes
Your context budget is tight; this one payload is heavier than most servers we measured in this round
You mostly want pipeline status, in which case a CLI call or a webhook is cheaper than 13 tool descriptions
You do not use CircleCI, which sounds obvious until you notice how many CI-shaped MCP servers get installed speculatively
You want the vendor's supported path, which the banner says is the hosted MCP server or the CircleCI CLI MCP

The bill for your context window

ToolEst. tokensDescribed?Required params
run_rollback_pipeline~1,595yes (5115 chars)0
list_component_versions~1,427yes (4608 chars)0
get_job_test_results~1,372yes (3389 chars)0
get_build_failure_logs~1,274yes (2805 chars)0
run_pipeline~1,225yes (2613 chars)0
get_latest_pipeline_status~1,114yes (2511 chars)0
download_usage_api_data~1,046yes (2792 chars)0
find_flaky_tests~975yes (2312 chars)0

Its tools/list is 51736 bytes, about 12932 tokens at 4 chars per token, measured 2026-08-19. That is heavy for 13 tools, and the weight is not spread evenly: run_rollback_pipeline alone is roughly 1595 tokens of description and list_component_versions about 1427, while config_helper costs about 378. The two rollback and component tools together cost more than the four cheapest tools combined. Every one of the 13 declares zero required parameters, so the descriptions are doing the work that a schema normally would, which is why they run long. If your client supports per-tool filtering, keep the failure-log, test-result and status tools and drop the rollback and usage ones until you actually need them.

Setup reality

npx -y @circleci/mcp-server-circleci, stdio, no config file. When we started it under a clean env with no credentials it still came up: the harness recorded no env hints and no error, so nothing is enforced at boot. In real use you supply a CircleCI personal API token by env var per the docs; without it the tools will authenticate against nothing and fail at call time rather than at startup. Cold start was slow, mostly npm fetching the package. Note the resolved version in the deprecation warning is older than the version string the server reports over the protocol, which is its own small mess.

Questions people ask

Is @circleci/mcp-server-circleci deprecated?

Yes. When we booted it, the package printed a deprecation notice saying this self-hosted server will stop receiving updates, including security fixes, and directing users to CircleCI's hosted MCP server or the CircleCI CLI MCP. It still works; it is just no longer the supported path.

Does it need a CircleCI API token to start?

Not to start. We ran it with a clean environment and no credentials and it initialized and listed all 13 tools anyway. The token matters when a tool is actually called, so a successful boot tells you nothing about whether your auth is right.

Why is it so slow to come up?

Our measured start was 17537 ms, and most of that is npx downloading the package on a cold cache rather than the server initializing. Pin and pre-install it locally and the second start is not comparable to the first.

Which tools are worth keeping enabled?

get_build_failure_logs, get_job_test_results, get_latest_pipeline_status and find_flaky_tests cover the debugging loop most people install this for. The rollback, component version and usage tools carry the largest descriptions and the least frequent use, so they are the first to disable.

Other dev tools servers we started

ServerVendorBootsContext tax
Everything (test server)Anthropic (reference)~1,913 tokens
GitAnthropic (reference)~1,477 tokens
GitHub (archived npm server)Anthropic (archived)~3,964 tokens
GitLab (archived)Anthropic (archived)~0 tokens

How this page is made: the server is spawned as an unprivileged user with a clean environment and no credentials, then asked for its tools, resources and prompts over stdio. Token figures are estimates at four characters per token, not a tokenizer count. One run, one machine. Corrections: contact the desk.

← All measured MCP servers