Provides tools for interacting with Jenkins, enabling AI agents to manage items, builds, nodes, queues, views, plugins, artifacts, and execute Groovy scripts through the Jenkins API.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
Microtech fork of lanbaoshen/mcp-jenkins (MIT), a Model
Context Protocol server that lets an AI assistant read a Jenkins instance. Upstream version 3.6.0
(commit 1d8587e) is the baseline; this fork is read-only by default and has three opt-in write tools, all limited to an allowlist of release jobs: trigger_build, stop_build, cancel_queue_item.
It is the transport behind the microtech-jenkins plugin in the microtech-hub marketplace. End users
do not run it by hand: the plugin's .mcp.json starts it and supplies the credentials.
Area | Upstream | This fork | Why |
Other write tools ( | Present, hidden by | Deleted | A flag can be forgotten or regress; deleted code can not run |
Starting, stopping, cancelling |
|
| Release automation can start, and undo, the release jobs and nothing else |
| Present | Deleted | It is code execution on the Jenkins controller |
| Present | Deleted | Node configs carry agent connection details and are not needed |
Transports | stdio, SSE, streamable-http (binds | stdio only | Nothing is ever exposed on a network port |
URL / credentials | A request header could replace the URL while the server's credentials were still sent | Taken only from the server's own configuration | A caller could aim your token at a host of its choice |
| Auto-loaded from the working directory | Only with an explicit | A stray file can not silently change credentials |
Job and artifact paths | Inserted into URLs unchecked | Segments validated and URL-quoted; | Stops a crafted name reaching other URLs |
Scope | Whole instance | Optional | Limit reads to the folders you care about |
Output size | Unbounded | Console 5000 lines, 2000 chars per line, regex up to 200 chars, artifacts 5 MiB | One call can not flood a model's context |
| Raw | Secret-looking values replaced by | Config files can hold tokens |
File logging | Wrote the URL and username to | None | No credentials metadata on disk |
| Missing | Added | Build history of a job, newest first |
Dependencies | Lockfile with advisories | Refreshed lockfile |
--read-only is still accepted and ignored, so existing configurations keep working.
Related MCP server: MCP Jenkins Server
uvx --from git+https://github.com/microtechos/jenkins-mcp@<tag-or-sha> mcp-jenkins --transport stdioOption | Environment variable | Meaning |
|
| Jenkins base URL (required) |
|
| User. Give both username and password, or neither (anonymous) |
|
| API token (preferred) or password |
|
| Seconds per request, default 5 |
|
| TLS verification, default on |
|
| Register |
|
| Comma-separated exact job full names the three tools may act on. Empty = none |
|
| Comma-separated folder/job full names that may be read, e.g. |
| Load the settings above from a file (never auto-discovered) |
The allowlist gates every tool that takes a job fullname and filters get_all_items / query_items
(parent folders stay visible so the tree can be navigated). Views, nodes, the queue and plugins are
instance-level metadata and are not filtered: for a hard boundary also give the Jenkins user only the
permissions it needs (Overall/Read, Job/Read on the folders, Job/ExtendedRead only if get_item_config
is wanted).
Group | Tools |
Items |
|
Builds |
|
Artifacts |
|
Views / queue / nodes |
|
Plugins |
|
Trigger (opt-in) |
|
get_all_items and an unfiltered query_items return very large results on a big instance: prefer
query_items with fullname_pattern and folder_depth.
uv sync
uv run pytest
uv run --with ruff ruff check src testsKeep the fork read-only apart from the three allowlisted tools: a new tool must be a read tool, and nothing else that sends a non-GET request to Jenkins is added without a separate, explicit decision. Upstream changes are taken by diff or cherry-pick (the history is not shared). Release by tagging; the plugin pins a full commit SHA.
MIT, as upstream: see LICENSE. This is not an official Jenkins product.
This server cannot be deployed
Connect AI agents to CloudBees Unify: feature flags, CI/CD, release orchestration, and security
Connect AI assistants to your approved computer for files, commands, browser and desktop tasks.
Plan Salesforce deploys, open pull requests and trigger pipelines from your AI client.
Connect AI assistants to authorized Windows and macOS devices for files, commands, and management.
