Test Results MCP Server

An MCP (Model Context Protocol) server that lets an LLM assistant (such as Claude Desktop) analyze automated test results, search logs, and file Jira bugs with steps to reproduce.

Why

Triaging a failed test run usually means jumping between the results report, log files and Jira. This server exposes those as tools, so an assistant can chain them: find the failures, group them by cause, look for evidence in the logs, and draft a bug for a human to approve.

Related MCP server: CI Investigator MCP

Tools

Quick start

Requires Python 3.10+.

git clone https://github.com/AnittaSusanThomas/test-results-mcp-server.git
cd test-results-mcp-server
python -m pip install -r requirements.txt

# run the tests
python -m pytest

# try the server in the MCP Inspector
npx @modelcontextprotocol/inspector python test_results_mcp_server.py

The server reads sample/results.xml and sample/logs/ by default. Point it at your own JUnit-style XML and log folder with the RESULTS_FILE and LOG_DIR environment variables.

Use it with Claude Desktop

Open Settings → Developer → Edit Config and add this to claude_desktop_config.json (use your own absolute paths):

{
  "mcpServers": {
    "test-results": {
      "command": "python",
      "args": ["C:/path/to/test_results_mcp_server.py"],
      "env": {
        "RESULTS_FILE": "C:/path/to/sample/results.xml",
        "LOG_DIR": "C:/path/to/sample/logs"
      }
    }
  }
}

Fully quit and reopen Claude Desktop, then try:

Summarize my latest test run, group the failures by cause, and search the logs for timeouts.

Optional: Jira

Add these to the same env block (see .env.example): JIRA_BASE_URL, JIRA_EMAIL, JIRA_API_TOKEN, JIRA_PROJECT_KEY, and optionally JIRA_ISSUE_TYPE. Create an API token at https://id.atlassian.com/manage-profile/security/api-tokens.

Design decisions

• Dry-run by default. create_jira_bug only previews the ticket unless dry_run=False is passed, so an assistant cannot create tickets by accident. The client also asks for approval on each tool call.

• Secrets stay out of code. Credentials come from environment variables.

• Restricted log access. search_logs only reads files inside LOG_DIR.

• Jira projects differ. The tool asks Jira which issue types the project allows and picks a valid one (preferred type, then Bug, then Task) instead of hardcoding "Bug".

• stdio transport. Suits a local server launched by the client. A shared deployment would use HTTP with authentication.

Limitations and next steps

• Reads a local JUnit XML file; it is not yet wired to a CI system.

• No duplicate-ticket check before creating a Jira issue.

• Planned: a run_tests tool that triggers a Jenkins job, and a compare_runs tool that shows new failures versus the previous run.

• Tested with mcp 1.x (mcp[cli]<2). MCP 2.x renamed FastMCP to MCPServer.

Project layout

test_results_mcp_server.py   the server
sample/                      dummy results and logs for trying it out
tests/test_server.py         pytest tests
.env.example                 settings reference (no real values)

License

MIT