Skip to main content
Glama

Mutmut MCP

A Model Context Protocol (MCP) server for managing mutation testing with mutmut. This tool provides a set of programmatic APIs for running mutation tests, analyzing results, and improving test coverage in Python projects. It targets mutmut 3.x: mutants are named (mymodule.x_add__mutmut_1), reported per mutant with a status (killed / survived / no tests / timeout / suspicious / skipped / segfault), and configured through [tool.mutmut] in pyproject.toml.

Features

  • Run mutation testing sessions against the projects mutmut 3.x is configured for (source_paths in [tool.mutmut]), not per-target CLI flags

  • Show results per mutant name and status, and list surviving mutations, including a labelled section for mutants no test covers

  • Show the code diff for a single mutant and rerun one mutant, all survivors, or the survivors no test covers

  • Rank survivors and uncovered mutants by likely materiality so an agent knows what to look at first

  • Clean mutmut's per-project state (mutants/ and the legacy cache) without touching other projects

  • Serialize concurrent runs: two mutmut operations against the same project never race; the second gets a structured busy error

  • Classify every failure by its explicit channel — non-zero exit, exception, or local validation — never by sniffing output text

  • Practices what it preaches: this repository's own CI runs the server against itself with source_paths = ["mutmut_mcp.py"] and fails on mutation-strength regressions (see How this server tests itself)

Related MCP server: MCP Python Server

Install

# Run directly from GitHub (no install needed)
uvx --from git+https://github.com/wdm0006/mutmut-mcp mutmut-mcp

# Or install from source
git clone https://github.com/wdm0006/mutmut-mcp
cd mutmut-mcp
uv sync
uv run mutmut_mcp.py

Installing via Smithery

To install mutmut-mcp for Claude Desktop automatically via Smithery:

npx -y @smithery/cli install @wdm0006/mutmut-mcp --client claude

MCP Client Configuration

{
  "mcpServers": {
    "mutmut": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/wdm0006/mutmut-mcp", "mutmut-mcp"]
    }
  }
}

API / Tools

The following tools are available:

  • run_mutmut(target, options="", venv_path=None, project_path=None) – Run a mutation testing session (mutmut run) for the project; survivors are expected and are a result, not a failure

  • start_mutmut_run(target="", options="", venv_path=None, project_path=None) – Start mutmut run in a background thread and return {run_id, state} immediately; holds the project's write lock until it ends (other write tools return the busy error)

  • mutmut_run_status(run_id) – {state: running|finished|failed, elapsed_s, exit_failed, output_tail, progress} for a background run; unknown ids return an error (the newest 20 jobs are kept, in memory only)

  • cancel_mutmut_run(run_id) – Terminate a background run (SIGTERM, then SIGKILL), free the project lock and mark it cancelled; mutmut state may be partial afterwards. Finished or unknown ids are a no-op with a message

  • show_results(venv_path=None, project_path=None) – Show the last campaign's results per mutant name and status

  • show_survivors(venv_path=None, project_path=None) – List surviving mutations, plus a labelled section for mutants no test covers

  • rerun_mutmut_on_survivor(mutation_id=None, venv_path=None, project_path=None) – Rerun a specific mutant (mutmut run <name>) or all surviving mutants

  • clean_mutmut_cache(venv_path=None, project_path=None) – Remove mutmut's mutants/ state directory and legacy cache for the project

  • show_mutant(mutation_id, venv_path=None, project_path=None) – Show the code diff and details for one named mutant

  • mutant_context(mutation_id, venv_path=None, project_path=None) – {mutant_id, diff, tests, message}: the mutant's diff plus the pytest ids covering it (mutmut show + mutmut tests-for-mutant); an empty tests list means no test covers it

  • prioritize_survivors(venv_path=None, project_path=None) – Rank survivors and uncovered mutants by likely materiality, highest score first

  • mutation_score(venv_path=None, project_path=None) – Killed/total counts and the mutation score (null when there are no results or the run is incomplete) from one mutmut results --all true read

project_path

Every tool accepts an optional project_path: the directory containing the project's mutmut configuration ([tool.mutmut] in pyproject.toml), its source, its tests, and mutmut's mutants/ state directory. Mutmut runs with that directory as its working directory, and clean_mutmut_cache removes state only from inside it.

Lock file

run_mutmut, start_mutmut_run, rerun_mutmut_on_survivor and clean_mutmut_cache take a cross-process advisory lock on <project_path>/.mutmut-mcp.lock (created on first use, kept outside mutants/ so cleaning does not delete it). A second MCP server, or any other process holding that lock, makes these tools return the "already in progress" error instead of corrupting mutmut's state; a plain terminal mutmut run does not take the lock. Read tools never lock. If the file cannot be created (read-only project) the server falls back to its in-process lock only. Add .mutmut-mcp.lock to your project's .gitignore.

Pass it whenever the server was not launched from the project directory — which is typical for desktop MCP clients and uvx. Omitted, the tools fall back to the server process's working directory, so existing calls behave as before. A project_path that is not an existing directory returns an error without running mutmut or deleting anything.

venv_path

venv_path points at the virtual environment holding the project's mutmut (the tools use <venv>/bin/mutmut, or <venv>\Scripts\mutmut.exe on Windows); omitted, mutmut is taken from PATH. A relative venv_path is resolved against project_path when both are given, so the common .venv form works:

{ "project_path": "/home/me/src/myproject", "venv_path": ".venv" }

How this server tests itself

The repository's own [tool.mutmut] configuration points mutmut at the server module (source_paths = ["mutmut_mcp.py"]), so CI runs the mutation-testing server against itself. The campaign result is scored by scripts/mutation_budget.py, which reuses the server's own strict results parser and fails the build when the number of undetected mutants (survived or no tests) exceeds a fixed budget. The six allowed survivors are equivalent mutants, each documented in docs/mutation-waivers.md; anything new is a regression and must be killed by a named test or waived with a written reason.

Locally, the same loop is available via make mutate; make coverage enforces the measured 100% line-coverage floor, and make typecheck runs mypy.

License

This project is licensed under the MIT License. See LICENSE for details.

Available Tools

7 tools
clean_mutmut_cacheB

Clean mutmut cache using the mutmut CLI (if available), otherwise remove .mutmut-cache file. Returns the plain text output or confirmation message.

Inputs: venv_path. Returns: result

prioritize_survivorsA

Prioritize surviving mutants by likely materiality, filtering out log/debug-only changes and ranking by potential impact. Returns a sorted list of survivors with reasons for prioritization.

Inputs: venv_path. Returns: object

rerun_mutmut_on_survivorB

Rerun mutmut on specific surviving mutations or all survivors after test updates using the mutmut CLI. Returns the plain text output.

Inputs: mutation_id, venv_path. Returns: result

run_mutmutA

Run a full mutation testing session with mutmut on the specified target.

This tool initiates mutation testing on the given module or package. You can provide additional mutmut options as needed. The output includes a summary of mutations tested, including counts of killed, survived, and timed-out mutations. If a virtual environment path is provided, mutmut will be run using the binaries from that environment to ensure compatibility with project-specific dependencies.

Args: target (str): The module or package to run mutation testing on. options (str): Additional command-line options for mutmut (e.g., '--use-coverage'). Defaults to empty. venv_path (Optional[str]): Path to the project's virtual environment to use for running mutmut. Defaults to None.

Returns: str: Summary of the mutation testing run, or error message if the run fails.

Inputs: target (required), options, venv_path. Returns: result

show_mutantB

Show the code diff and details for a specific mutant using mutmut show. Args: mutation_id (str): The ID of the mutant to show. venv_path (Optional[str]): Path to the virtual environment, if any. Returns: str: The output of 'mutmut show <mutation_id>'.

Inputs: mutation_id (required), venv_path. Returns: result

show_resultsC

Display overall results from the last mutmut run using the mutmut CLI. Returns the plain text output.

Inputs: venv_path. Returns: result

show_survivorsC

List details of surviving mutations from the last mutmut run using the mutmut CLI. Returns the plain text output.

Inputs: venv_path. Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.7/5.0

    Scored across 7 tools

    Disambiguation5/5

    Each tool has a distinct purpose: run_mutmut initiates testing, show_results displays overall results, show_survivors lists survivors, show_mutant shows specific mutant details, rerun_mutmut_on_survivor retests survivors, prioritize_survivors ranks survivors, and clean_mutmut_cache manages cache. There is no overlap or ambiguity in functionality.

    Naming Consistency5/5

    All tools follow a consistent snake_case naming pattern with clear verb_noun structures (e.g., run_mutmut, show_results, clean_mutmut_cache). The naming is predictable and readable throughout the set.

    Tool Count5/5

    With 7 tools, this server is well-scoped for mutation testing with mutmut. Each tool serves a specific role in the workflow, from running tests to analyzing results, without being overly sparse or bloated.

    Completeness5/5

    The tool set provides complete coverage for the mutmut domain: it includes core testing (run_mutmut), result inspection (show_results, show_survivors, show_mutant), survivor management (rerun_mutmut_on_survivor, prioritize_survivors), and maintenance (clean_mutmut_cache). There are no obvious gaps in the mutation testing lifecycle.

    Maintenance

    ActivityActive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      F
      maintenance
      A Model Context Protocol server that enables AI assistants like Claude to perform Python development tasks through file operations, code analysis, project management, and safe code execution.
      9
      MIT
    • F
      license
      Not graded
      quality
      D
      maintenance
      A Python-based implementation of the Model Context Protocol that enables communication between a model context management server and client through a request-response architecture.
      -
    • F
      license
      A
      quality
      Not graded
      maintenance
      A Model Context Protocol server that allows LLMs to interact with Python environments, enabling code execution, file operations, package management, and development workflows.
      9
      -