MCP server
Semaloop’s MCP server lets a coding agent do anything the dashboard can: create and update tests from a natural language description, trigger runs, and pull the video, logs and root cause behind a result.
Configuration and capabilities
Section titled “Configuration and capabilities”Our MCP server allows you to interact with Semaloop directly from your coding agent. This means that you no longer need to log in to the Semaloop dashboard to manage any aspect of your account, as it’s all possible with MCP.
To get started, go to Settings → MCP.
To get the latest list of what’s possible, just ask Semaloop MCP: “What tools are available in the Semaloop MCP server”:
- Discover your apps, builds, and tests: See what exists before acting on it.
- Create and update tests: Turn a natural language description of a user journey into a test, and revise it as the app changes.
- Trigger runs and access their context: Kick off a run and pull everything needed to diagnose it, including root-cause analysis, video recording, network logs, and system logs.
- Track recurring issues: See issues that show up repeatedly across runs, not just one-off failures.
- Create and update secrets: Manage the credentials and values a run needs to sign in or fill in forms.
- Create and update custom tools: Define HTTP calls the run agent can use for things outside the UI, like seeding data or resetting an account.
- Send feedback: Flag anything the agent needed to do but couldn’t.
Connecting with an API key
Section titled “Connecting with an API key”Signing in through the browser suits a coding agent on your own machine. A
service, script or CI job has no browser to sign in with, so it can connect
with an API key instead. Create one in
Settings → API keys, then point the client at
https://mcp.semaloop.com/mcp/key and send the key as a bearer token on every
request. For example, with Claude Code:
claude mcp add --transport http semaloop https://mcp.semaloop.com/mcp/key \ --header "Authorization: Bearer <api-key>"Any MCP client that supports Streamable HTTP and custom headers works the same way. The tools and prompts are the same as when you sign in, with two differences:
- One app per key. An API key belongs to a single app, so the session can only see and act on that app. To work across several apps, use a key for each.
- No confirmation prompts. The key is the authorisation, so tools that make changes, such as creating a test or cancelling a run, run without the client asking first. Treat the key like any other credential. Disabling it in Settings → API keys cuts off access on the next request.
Example use cases
Section titled “Example use cases”Here are a couple of recommendations on how you can leverage our MCP server:
- Generating tests from a PR description. Point your agent at an open PR and ask it to check whether Semaloop already covers the change. It can list your existing tests, compare them against the PR description, and if the flow isn’t covered, write a new test directly from that description. This means that your coverage grows with the codebase, with no one having to remember to open the dashboard.
- Fixing failures on the spot. When a run fails, ask your agent to pull the run, video, network logs, and system logs, then work out whether the app broke or the test itself was brittle. It can push a code fix, or update the test spec, all within the same session. See the debugging failures documentation.
- Pushing recurring issues into your issue tracker. Rather than reacting to every issue individually, ask your agent to pull the issues for your app and combine them with other details that you have in your organisation, like meeting notes, support tickets, and other customer context.