Jenkins MCP server: 18 tools from a plugin

Jenkins becomes the server rather than connecting to one. What the eighteen tools reach, how log pagination works, and the four tools that change a build.

Last verified 1 September 2026 · from The Jenkins plugin repository and its own README

This page is one of 90 in a directory of monitoring and developer tools. Each page checks what an AI coding agent can get out of the tool through its MCP server, the connector that lets the agent query the tool directly.

What is the Jenkins MCP server?

It is an MCP server implemented as a Jenkins plugin, exposing the controller over the Model Context Protocol. Because it runs inside Jenkins, there is no separate process to deploy.

The tool set follows the Jenkins object model: jobs, builds, the queue, test results and SCM configuration. getJobs pages through jobs by name and getBuild fetches a specific build or the last one.

Four tools go beyond reading. triggerBuild starts a build, and updateBuild changes a build's display name or description. rebuildBuild re-runs a build. replayBuild re-runs a Pipeline with a modified script.

Jenkins is a CI system, so it sits in CI and release, touches code and tests through test results, and reaches nothing in agent observability or cost.

In practice, Jenkins answers what ran, what it did and why it failed, and can start it again. It has no view of the running service, so a question about production behavior needs a telemetry connector beside it.

What can an AI agent do with Jenkins?

getBuildLog returns a nextCursor on every response. Pass it back to continue the read without re-scanning. That is what makes a long log usable inside a context window.

What connecting costs before the first question

A context window is the amount of text a model can hold at once. Eighteen tools load when the plugin is enabled. They cost 1,450 to 1,604 tokens on Claude. Jenkins is the server here rather than a client of one, so the definitions come from your own controller.

Four of the eighteen change a build. The context cost is the same whether you intend to use them or not, because the plugin registers the full set.

Encodings: GPT-5.6, GPT-5.5 o200k_base; GPT-4, GPT-3.5 cl100k_base; Claude Haiku 4.5 to Opus 5 count_tokens. Counted from the 18 tool entries in the vendor's own documentation. That is the only public inventory.

The server could not be run for a live tools/list. So no input schemas are available. These figures cover each tool's name and description only. They are a floor, not the figure.

The Claude columns are measured through the API's count_tokens endpoint, with an empty schema per tool. So they cover the same content as the GPT columns. The range there is two tokenizer generations. Haiku 4.5 and Opus 4.6 sit at the low end. Opus 5 and Sonnet 5 sit at the high end. The gap between them is about a tenth.

Jenkins MCP server limits: where answers come back incomplete

The limits below come from logs being large and from the server living inside the thing it describes.

How to configure Jenkins MCP for agents

1. 1

   Decide who may trigger builds

   triggerBuild, rebuildBuild and replayBuild all start work, and replayBuild runs a modified Pipeline script.

2. 2

   Prefer streamable HTTP

   The README recommends it. Streamable HTTP answers one request with one response, so it needs none of the keep-alive handling that server-sent events require.

3. 3

   Leave Origin validation on

   The specification marks validating the Origin header as a MUST, and the plugin validates it against the configured Jenkins root URL by default.

4. 4

   Know what the health endpoint exposes

   /mcp-health returns server status and active connection counts and, by design, requires no authentication.

Paste this into your agent’s instructions

Log reads are paginated and cursor-based, so a partial read looks like a complete one. The block makes the agent say which it did.

When you answer using Jenkins MCP tools, state:
- The job and build number the answer covers.
- Whether a log read was complete. getBuildLog paginates, and
  totalLines is -1 for forward and cursor reads.
- Whether you called triggerBuild, rebuildBuild, replayBuild
  or updateBuild. All four change state.
Never state that data does not exist. State that the build
above returned no matching lines or results.
Write your answer in ASD-STE100 Simplified Technical
English. Use short sentences with one idea in each.

Do you need the MCP server at all?

Not for a scripted trigger. Jenkins has a REST API and a CLI, and for that either is simpler than a connector.

The connector earns its place on log reading. Cursor pagination and end-relative reads are exactly what a model needs to work through a long build log without exhausting its window.

For test analytics across runs rather than one build, ReportPortal or BuildPulse answer questions this does not.

Connect it when an agent is debugging a build. Use the REST API for scripted triggers.

Jenkins MCP server setup

Install the MCP Server plugin through the Jenkins plugin manager. The README states that it sets up the necessary endpoints and tools on installation.

Clients connect to the controller itself. The streamable HTTP endpoint is at /mcp-server/mcp under your Jenkins URL.

claude mcp add --transport http jenkins \
  https://jenkins.internal/mcp-server/mcp \
  --header "Authorization: Basic <user:token base64>"

{"mcpServers":{"jenkins":{
  "url":"https://jenkins.internal/mcp-server/mcp",
  "headers":{"Authorization":"Basic <user:token base64>"}}}}

[mcp_servers.jenkins]
url = "https://jenkins.internal/mcp-server/mcp"
headers = { Authorization = "Basic <user:token base64>" }

Every other client

Each block below is the configuration for one client, with the file path and the key that client expects.

{"mcpServers":{"jenkins":{"url":"https://jenkins.internal/mcp-server/mcp",
  "headers":{"Authorization":"Basic <user:token base64>"}}}}

{"mcpServers":{"jenkins":{"url":"https://jenkins.internal/mcp-server/mcp",
  "headers":{"Authorization":"Basic <user:token base64>"}}}}

{"servers":{"jenkins":{"url":"https://jenkins.internal/mcp-server/mcp",
  "headers":{"Authorization":"Basic <user:token base64>"}}}}

{"mcpServers":{"jenkins":{"url":"https://jenkins.internal/mcp-server/mcp",
  "headers":{"Authorization":"Basic <user:token base64>"}}}}

-Dio.jenkins.plugins.mcp.server.Endpoint\
.disableMcpStreamable=true

curl https://jenkins.internal/mcp-health

Read-only access, permissions and security

• Four tools change state

  triggerBuild, rebuildBuild, replayBuild and updateBuild start or alter builds.

• replayBuild runs a modified script

  It re-runs a Pipeline with a script the caller supplies, which is code execution on your controller.

• The health endpoint is unauthenticated by design

  /mcp-health returns status and active connection counts without a credential. The plugin does this by design, so keep the endpoint inside your monitoring network.

• Origin validation is on by default

  The plugin validates the Origin header against the configured Jenkins root URL, which the MCP specification requires.

Troubleshooting

A cursor is rejected

Cursors are tied to the job and build number they were issued for, and are rejected against a different build.

A long tool call returns 504

A slow triggerBuild can exceed a short proxy read timeout. The README notes raising proxy_read_timeout for that case, regardless of transport.

Total line count looks wrong

totalLines is exact only for end-relative reads. Forward and cursor reads return -1, because they stop as soon as they have enough lines.

Jenkins MCP server: Reference

What engineers report

jenkinsci/mcp-server-plugin carries 107 stars, 64 forks and 46 releases since 27 May 2025. The README is detailed about transports and pagination, so the counts below come from it.

Repository facts from the GitHub API, retrieved 24 August 2026. Tool inventory, SDK version and transport behavior read from the plugin's own README on the same date. The server runs only inside a Jenkins controller, so no standalone tools/list capture was possible.

Should you connect an agent to Jenkins?

Yes. Best for streamable HTTP with Origin validation left on, and client approval required for the four acting tools.

Strongest fit

Teams running their own Jenkins who want an agent to read build logs without a browser.

Main advantage

Cursor-based log pagination built for a context window, from a server that needs no separate deployment.

Main weakness

It describes Jenkins only, so anything about the deployed service needs another connector.

Operational risk

replayBuild executes a caller-supplied Pipeline script on the controller.

Frequently asked questions