Diffblue MCP server: three tools over the dcover CLI

One of the smallest servers in this guide, what its three tools actually invoke, and the two prerequisites that decide whether it runs at all.

Last verified 1 September 2026 · from Diffblue's repository and the server's own tools/list response

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 Diffblue MCP server?

It is Diffblue's server for the Model Context Protocol. Diffblue Cover writes JUnit tests for Java automatically, so a connected agent can ask for tests rather than writing them itself.

The server is deliberately thin. Its own README calls the script a universal adapter for the dcover create command, and the three tool descriptions each say they execute a dcover subcommand.

That makes the tool list a faithful map of the CLI. create generates tests and refactor applies automated fixes, aliased as fix-build.

issues returns a prioritized list of problems that may prevent test generation, which is why it is worth calling first.

Diffblue sits in code, tests and review through a three-tool wrapper around its Cover command line, and reaches nothing in CI, production observability, agent observability or cost.

For an agent, that means Diffblue answers one question: can this Java code have tests written for it, and what is stopping that. It has no view of the build, the pipeline or production, so it sits alongside those connectors rather than replacing any of them.

What can an AI agent do with Diffblue?

Run issues before create. Its description exists precisely because generation fails for reasons that are worth reading first.

What connecting costs before the first question

A context window is the amount of text a model can hold at once, and connecting uses some of it before the agent asks anything. Every tool the server registers sends its name, description and input schema to the model when the session starts. The tool list is charged whether or not a single tool is called.

3 tools cost between 836 and 1,817 tokens, or 0.4% to 0.9% of a 200,000-token window.

The schemas carry it. Descriptions average 220 characters and schemas average 1,004, because each tool passes a long list of CLI options through to dcover.

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 24 August 2026 from tools/list on the package installed from the git repository and run under Python 3.14.

Registration is static, so no Diffblue license was involved in the capture. Figures cover name, description and input schema. The two models in column one share the o200k_base encoding, so they produce one count. The 66% spread inside a column is formatting, compact to indented. Reasoning tokens are output, spent per request. The tool list is input, charged once at connection.

The Claude figures are measured rather than encoded. Anthropic publishes no local tokenizer, so the definitions were sent to the API's count_tokens endpoint and a no-tools baseline subtracted. The range covers two tokenizer generations rather than two serializations. Haiku 4.5 and Opus 4.6 sit at the low end, Opus 5 and Sonnet 5 at the high end. The gap between them is about a tenth.

Diffblue MCP server limits: where answers come back incomplete

Both limits below come from the server being an adapter rather than an implementation.

How to configure Diffblue MCP for agents

1. 1

   Check your Python version first

   The package declares requires-python >= 3.14, so an older interpreter refuses to install it.

2. 2

   Install the CLI before the server

   dcover must be on your PATH; the README suggests verifying with dcover version.

3. 3

   Know that two tools write

   create adds test files and refactor changes project files, both through the CLI.

4. 4

   Pin your own clone

   With nothing on PyPI, the version you run is whatever the repository held when you installed it.

Paste this into your agent’s instructions

Every tool here shells out, so a failure can come from the server, the CLI or the license. The block makes the agent say which.

When you answer using Diffblue MCP tools, state:
- The dcover subcommand that ran, and the path it ran against.
- Whether files were written. create adds tests and refactor
  changes project files.
- Whether a failure came from the MCP server or from the
  dcover CLI it invoked.
Never state that data does not exist. State that the command
above produced no output.
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?

Usually not. dcover create in a terminal does exactly what the create tool does, without a server in between.

The connector earns its place when the agent is already editing Java and can call generation as part of the same conversation, then read issues when it fails.

For test quality signals rather than test generation, SonarQube and Codacy report coverage and duplication that this connector does not touch.

Use the CLI directly for batch work. Connect the server when an agent is writing the Java it wants tested.

Diffblue MCP server setup

Two prerequisites come before any client configuration: the dcover CLI on your PATH, and Python 3.14 or later.

The package is not on PyPI, so install it from the git repository. It exposes no console script; the server is a module.

claude mcp add-json "diffblue" '{"command":"python3.14",
  "args":["-c","from covermcp.server import mcp; mcp.run()"]}'

{"mcpServers":{"diffblue":{"command":"python3.14",
  "args":["-c","from covermcp.server import mcp; mcp.run()"]}}}

[mcp_servers.diffblue]
command = "python3.14"
args = ["-c", "from covermcp.server import mcp; mcp.run()"]

Every other client

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

{"mcpServers":{"diffblue":{"command":"python3.14",
  "args":["-c","from covermcp.server import mcp; mcp.run()"]}}}

{"mcpServers":{"diffblue":{"command":"python3.14",
  "args":["-c","from covermcp.server import mcp; mcp.run()"]}}}

{"servers":{"diffblue":{"command":"python3.14",
  "args":["-c","from covermcp.server import mcp; mcp.run()"]}}}

{"mcpServers":{"diffblue":{"command":"python3.14",
  "args":["-c","from covermcp.server import mcp; mcp.run()"]}}}

python3.14 -m pip install \
  "git+https://github.com/diffblue/cover-mcp.git"
# then verify the CLI it wraps:
dcover version

Read-only access, permissions and security

• Every tool executes a local command

  The server's job is to invoke dcover with arguments the model supplies.

• Two of the three write to your working tree

  create adds test files and refactor applies automated changes to project files.

• The license boundary is the CLI's

  The server holds no credential; whatever dcover is entitled to do on that machine is what the agent can trigger.

• There is no pinned version

  With no PyPI release, the installed code is whatever the default branch held at install time.

Troubleshooting

Installation is refused

The package declares requires-python >= 3.14. Older interpreters, including 3.12, are rejected before anything is installed.

No console script appears after installing

The package ships no entry point. Start it as a module instead, importing mcp from covermcp.server and calling run().

Every tool call fails

dcover is probably not on the PATH. The README suggests confirming with dcover version before configuring any client.

Test generation produces nothing

Call issues first. It returns a prioritized list of project problems that may prevent test generation.

Diffblue MCP server: Reference

What engineers report

Adoption is small and the repository is quiet, which is the honest summary of a three-tool adapter.

Repository facts from the GitHub API, retrieved 24 August 2026. PyPI was checked for cover-mcp on the same date and returned 404. Tool counts and sizes come from the server's own tools/list response on the installed package.

Should you connect an agent to Diffblue?

Yes, though there is no packaged release. Best for a dedicated 3.14 interpreter, a verified dcover install and a clean branch to run against.

Strongest fit

Java teams that already license Diffblue Cover and want test generation inside an agent conversation.

Main advantage

A faithful, tiny surface: three tools that map exactly onto the CLI subcommands they invoke.

Main weakness

Nothing is published, and the package needs Python 3.14, which most machines do not have.

Operational risk

Two tools write to the working tree by invoking a local CLI with model-supplied arguments.

Frequently asked questions