Run your TestLink test management from Claude — in plain English. Create suites, write and refine test cases, build test plans, record executions, and link requirements, all through natural-language chat with any Model Context Protocol client. No more clicking through the TestLink UI.
TestLink is powerful but slow to drive by hand — every case edit, plan, and execution is a trip through the web UI. testlink-mcp puts a clean MCP server in front of TestLink's XML-RPC API so an AI assistant can do that work for you:
- Talk, don't click. "Add detailed steps to MFT-12 and mark it high importance."
- Works with any MCP client — Claude Code, Cursor, Dify, or your own.
- 30 tools across the whole TestLink model — cases, suites, plans, builds, executions, requirements, projects.
- Solid where it counts — strict input validation, clear XML-RPC error mapping, and correct internal-vs-external ID handling so edits land on the right case.
- One container to run it.
docker runand you're connected.
You ask in plain language. Your MCP client (Claude) picks the right tool and arguments; testlink-mcp validates them and makes the matching XML-RPC call to TestLink; the result comes back as a plain-language answer. The server speaks MCP over stdio and TestLink over XML-RPC — you only ever see the conversation.
The published Docker image is the fastest way in.
Claude Code
claude mcp add testlink -- docker run --rm -i \
-e TESTLINK_URL=http://your-testlink-host/testlink \
-e TESTLINK_API_KEY=your_api_key_here \
dogkeeper886/testlink-mcp:latestCursor / other MCP clients — add to your MCP config:
{
"mcpServers": {
"testlink": {
"command": "docker",
"args": ["run", "--rm", "-i",
"-e", "TESTLINK_URL=http://your-testlink-host/testlink",
"-e", "TESTLINK_API_KEY=your_api_key_here",
"dogkeeper886/testlink-mcp:latest"]
}
}
}Then ask your assistant: "List the test suites in project 1."
Get your API key: in TestLink, My Settings → API interface → Generate key, and make sure the XML-RPC API is enabled for your installation.
testlink-mcp covers the full TestLink object model — the requirement that covers a case, the case that lives in a suite and a plan, the build that belongs to a plan, and the execution that records a result.
| Area | Tools |
|---|---|
| Test Cases (4) | create_test_case, read_test_case, update_test_case, delete_test_case |
| Test Suites (5) | create_test_suite, update_test_suite, delete_test_suite, list_test_suites, list_test_cases_in_suite |
| Test Plans (5) | create_test_plan, delete_test_plan, list_test_plans, add_test_case_to_test_plan, get_test_cases_for_test_plan |
| Builds (3) | create_build, close_build, list_builds |
| Executions (2) | create_test_execution, read_test_execution |
| Requirements (7) | create_requirement_specification, delete_requirement_specification, list_requirement_specifications, create_requirement, get_requirement, list_requirements, assign_requirements |
| Projects (4) | create_project, update_project, delete_project, list_projects |
Every tool above works against a stock TestLink with the XML-RPC API enabled, with one
exception: update_project needs tl.updateTestProject, a method upstream TestLink
does not ship. It is available in the
dogkeeper886/testlink-code fork on the
main branch. Point update_project at a server without it and the tool says so by
name rather than returning a raw XML-RPC fault.
On top of the raw tools, the repo ships three Claude skills — packaged workflows your assistant invokes when you ask:
| Skill | What it does |
|---|---|
testlink-sync |
Turns a source document — a spec, a GitHub issue, a folder of markdown test cases — into TestLink content: requirements, suites, cases, plans, builds. |
testlink-review |
Reads the written content back and verifies it met your goal — the right entities exist, fields are correct and well-formed. |
testlink-format |
The HTML markup reference for TestLink's rich-text fields, used by the other two. |
To use them: copy the skill folders from .claude/skills/ into your
project's .claude/skills/ (or ~/.claude/skills/ for every project), then just ask —
"Sync the test cases in docs/tests/ into TestLink, then review them."
| Variable | Required | Description |
|---|---|---|
TESTLINK_URL |
yes | Base URL of your TestLink instance (e.g. http://host/testlink) |
TESTLINK_API_KEY |
yes | Your TestLink API key (My Settings → API interface) |
The server exits immediately if TESTLINK_API_KEY is missing.
Pass them as environment variables, or put them in a .env file — the server loads
one automatically at startup:
cp .env.example .env # then edit; .env is gitignoredgit clone https://github.com/dogkeeper886/testlink-mcp.git
cd testlink-mcp
npm install
npm run build
TESTLINK_URL=http://host/testlink TESTLINK_API_KEY=key node dist/index.jsYou need a reachable TestLink with the XML-RPC API enabled. If you don't have one, the
testlink-code fork ships a disposable
stack:
cicd/scripts/ci-up.sh # build, start, seed — TestLink on :8091
cicd/scripts/ci-down.sh # tear downIt is ephemeral — the fork's test runner tears it down when it finishes, so expect to
start it again. Point your .env at http://localhost:8091.
- "Cannot connect to TestLink" — check
TESTLINK_URLis reachable from the container and that the XML-RPC API is enabled. - "Invalid API key" — regenerate the key in TestLink and confirm API access is on for your user.
- Edits hit the wrong case — pass the external ID (
PREFIX-123) or the numeric internal ID; the server routes each correctly.
testlink-mcp/
├── src/index.ts # the MCP server (tools, handlers, XML-RPC client)
├── cicd/tests/ # end-to-end test runner + the dual-judge framework
├── .claude/ # dev-workflow + qa-workflow commands and skills
├── docs/ # diagrams, stories, release notes
├── Dockerfile # multi-stage node:20-alpine build
└── .env.example # environment template
Issues and PRs welcome. The repo follows a gated dev-workflow (.claude/commands/);
the integration suite under cicd/tests must stay green.
No license is published yet — please open an issue if you need licensing clarified before use.
Questions and bugs: GitHub issues.