Pitest MCP server: a community server for mutation results

What the two community servers read, why neither can start a mutation run, and how to give them the XML report they need.

Checked 3 October 2026 · from pitest.org, the hcoles GitHub organization, the MCP registry, the GitHub API and the NaelDj/PIT-MCP-Server README

One of 91 guides to monitoring and developer tools. Each covers what an AI coding agent can get out of the tool through MCP, the protocol that lets the agent query the tool directly.

What is the Pitest (PIT) MCP server?

PIT is a mutation testing tool for Java. It changes your code in small ways, one mutation at a time, then runs your tests to see whether they notice. A mutation nothing catches is a gap in the test suite.

The output is a mutation score and a list of the mutants that survived, by class, method and line. PIT writes HTML by default, and XML or CSV when its outputFormats setting asks for them.

A connector lets an agent ask which mutants survived in one method without pasting the whole report into the prompt. NaelDj/PIT-MCP-Server does that from the XML report, so the build must be set to write XML.

Where a PIT connector could come from, and what is there:

Both community rows read reports a build already wrote. Neither can start PIT, and neither sees a run that has not finished.

Mutation testing sits in code, tests and review, and reaches it only through a file.

PIT results reach an agent only after a build writes a report and a community server parses it. A question cannot trigger a fresh run.

What can an AI agent do with Pitest (PIT)?

Agents reach PIT results only through community servers. NaelDj/PIT-MCP-Server answers per class, per method and surviving-mutant questions from the last mutations.xml, so an agent can ask about a finished run and cannot start one. The mutation run itself stays in the build.

Pitest (PIT) MCP server limits: where answers come back incomplete

Both PIT servers read reports. The limits below follow from that.

How to configure Pitest (PIT) MCP for agents

1. 1

   Turn on XML output

   PIT writes HTML by default. NaelDj/PIT-MCP-Server reads mutations.xml, so add XML to outputFormats in the Maven or Gradle plugin.

2. 2

   Do not expect to trigger a run

   Neither server starts PIT. That stays a build step, invoked by Maven, Gradle or your pipeline.

3. 3

   Weigh the dependency

   Both community servers have no stars, and NaelDj/PIT-MCP-Server ships no license file, so reuse terms are unclear.

4. 4

   Compare with reading the file

   An agent with file access can read target/pit-reports directly. The server helps most when the report is too large to paste.

Paste this into your agent’s instructions

A mutation score without its run is a number with no meaning. The block ties every figure to the report that produced it.

When you answer about mutation testing, state:
- The report file you read and when the run that produced it
  finished. A mutation score describes one run.
- Which classes the run covered. PIT is usually scoped to a
  package, not the whole codebase.
- That nothing here can start a run, so the newest figure
  available is the last build's.
Never state that data does not exist. State the report you
read and what it contained.
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 always. The report is a file, and an agent with file access can read target/pit-reports without any server.

PIT writes XML and CSV when outputFormats requests them, and both parse more reliably than a rendered HTML page.

Stryker Mutator covers the same idea for JavaScript, TypeScript, C# and Scala, and has a connector story of its own.

For test quality more broadly, SonarQube and BuildPulse both publish servers and answer coverage and flakiness questions directly.

Turn on XML output first. Then decide whether a server adds anything over reading mutations.xml directly.

Pitest (PIT) MCP server setup

PIT's maintainers have not published a server. NaelDj/PIT-MCP-Server is the community option with written setup steps, and it reads the mutations.xml that PIT has already written.

Set expectations before configuring anything. It cannot run mutation testing or reach PIT itself, so the mutation run stays in your build.

Its README covers the clone, a Python virtual environment and a Cline configuration. The steps below follow it.

<outputFormats>
  <param>XML</param>
  <param>HTML</param>
</outputFormats>
<!-- mvn org.pitest:pitest-maven:mutationCoverage
     writes target/pit-reports/mutations.xml -->

git clone https://github.com/NaelDj/PIT-MCP-Server.git
cd PIT-MCP-Server
python3 -m venv .venv && source .venv/bin/activate
pip install mcp anyio

{"mcpServers":{"PIT-MCP-Server":{
  "command":"/home/<user>/PIT-MCP-Server/.venv/bin/python",
  "args":["/home/<user>/PIT-MCP-Server/server.py"]}}}

Every other client

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

gh api repos/Barry2llen/TestReportHtmlAnalyzerMCP/contents \
  --jq '.[].name'    # no README

Read-only access, permissions and security

• Neither community server has stars or outside reviewers

  Adding one means running unreviewed code that parses build output.

• Reports describe your source

  A mutation report names classes, methods and line numbers, so it is a map of the codebase.

• Nothing writes

  Both servers read report files. They cannot change code, tests or a build.

• PIT itself modifies bytecode

  That happens inside the build, under your own control, and no connector changes it.

Troubleshooting

You cannot find an official PIT MCP server

The PIT project offers none, so pick a community server by report format: NaelDj/PIT-MCP-Server for XML, Barry2llen/TestReportHtmlAnalyzerMCP for HTML.

`pit_find_latest_xml` finds nothing

PIT wrote HTML only, which is its default. Add XML to outputFormats, rerun PIT, and point workspace at the project root.

You want the agent to run mutation testing

Neither server can. Invoke PIT through Maven or Gradle in the build, then read the report it produces.

Arcmutate features are missing

Arcmutate is the commercial extension set for PIT, and its sitemap contains no MCP references either.

Pitest (PIT) MCP server: Reference

What engineers report

PIT's maintainers have not published a connector, so the counts below cover the searches and the two community repositories.

“MCP server for querying PIT mutation testing results.”

“a stdio mcp server that analyzes html created by jacoco and pitest”

On 3 October 2026 the MCP registry returned zero results for pitest, a GitHub search of the hcoles organization for MCP repositories returned zero, and the Arcmutate sitemap contained no MCP references. pitest.org does not document an MCP server. A GitHub search for pitest and MCP returned the two community repositories named above, with facts from the GitHub API on the same date.

Should you connect an agent to Pitest (PIT)?

Only if a community server is acceptable to you. Best for java teams that run PIT in CI with XML output and want an agent to fix surviving mutants method by method.

Strongest fit

Yes, if your build writes mutations.xml and the report is too large to paste into a prompt. NaelDj/PIT-MCP-Server then lets an agent ask for surviving mutants one method at a time.

Main advantage

Scoped queries: per class, per method and surviving mutants, instead of the whole report in context.

Main weakness

It reads a finished report and cannot start PIT, so it adds no reach beyond the file.

Operational risk

Low. It reads files and writes nothing. The risk is depending on an unlicensed, unreviewed thesis project.

Frequently asked questions