Learn
What is MCP in software testing?
In software testing, MCP (the Model Context Protocol) is the open standard an AI agent such as Claude Code, Codex or Cursor uses to call a testing tool directly. A test management tool's MCP server exposes actions like reading test cases, creating them and recording results, so the agent works in the system of record instead of a chat window.
01MCP in one paragraph
The Model Context Protocol defines how an AI application (the client, such as Claude Code, Codex, Cursor or Claude Desktop) talks to a program that offers capabilities (the server). Messages are JSON-RPC 2.0, sent over standard input and output for a local server or over Streamable HTTP for a remote one. A server can offer tools (actions the model may call), resources (data the client can read) and prompts (templates). The client asks the server what it offers, and the model decides when to call each tool.
02MCP server vs API integration
| Classic API integration | MCP server | |
|---|---|---|
| Who calls it | Code written for one integration | Any MCP client, driven by a model |
| How actions are discovered | Read the API docs, write code | The client calls tools/list and gets names, descriptions and argument schemas |
| Who decides what to call | The developer, in advance | The model, at run time, from your instruction |
| Authentication | API key, OAuth or token per integration | The same, sent by the client; for example a Bearer key in a header |
| Adding a new client | New integration work | Point the client at the server URL |
| Good for | Fixed, repeatable pipelines such as CI uploads | Open-ended work: drafting cases, triaging a run, answering questions |
They complement each other. A CI pipeline should upload results through a stable API with a service key; an agent helping a person write or review cases is the case for MCP.
03What a test management MCP server exposes
A test management tool's MCP server turns the test repository into tools. TestOrchestrator's server, for example, offers 39 tools, which group into:
- Discovery: projects, folders, fields and templates, lifecycle states, statuses, tags, environments, cycles, users, connected trackers.
- Test cases: list, search by keyword, read, create and update, including steps and priority.
- Runs and results: start a run, record a result per case with a comment, read the run's tally, close it.
- Traceability: link a case to a Jira or Linear issue, find the cases for an issue, read coverage gaps.
- Comments, cycles, attachments and automation keys.
It offers tools only. When a client connects, the server's initialize answer advertises the tools capability; it publishes no resources or prompts.
04Where MCP fits in the testing workflow
- Authoring: the agent reads a ticket or the code it just changed and drafts cases in the right folder and template. See how to connect Claude Code.
- Running and recording: the agent opens a run, records what passed and failed with a comment, and leaves the close to a person.
- Tracing: the agent links each case to its issue and reports which issues have no cases yet.
- Reviewing: the agent summarises a run or lists draft cases waiting for review. See how to review test cases created by AI agents.
05Limits worth knowing
- The agent is only as good as its instruction. It will create what you ask for, in the state you ask for. Review its work.
- Search is what the server implements. TestOrchestrator's
search_test_casesmatches on title and case number, not meaning. - Request and response only. TestOrchestrator's server sends no server-initiated messages; the agent asks, the server answers.
- Permissions come from the key. An MCP server can only be as safe as the access its keys carry. Read security and permissions for MCP test management.