e2e MCP
An agent gets a live session of your web, iOS or Android app, sees the screen, writes tests and checks its own edits.
- 13,786 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The agent drives the app and performs actions in it inside the test environment.
- Everything visible on the session screen goes into the model context.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/e2eDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent drives the app and performs actions in it inside the test environment.
- Everything visible on the session screen goes into the model context.
This is third-party code. Review the repository files before installing.
What it does
The `e2e mcp` command starts the e2e platform MCP server and gives an agent such as Claude Code, Codex or Cursor a live session of your app on web, iOS or Android. The agent looks at the screen, acts on it and checks its own work. The server has five entry points: `open_session` opens a session with the app, `tools` shows the action catalog, `call` invokes a catalog tool, `close_session` closes the session and `run_tests` runs the project tests. The agent writes a test as it walks the flow from checked lines instead of guessing locators blind. The server needs no model: only an agent step without a recording does. The `e2e init` wizard asks whether to register the server in `.mcp.json` for Claude Code, `.cursor/mcp.json` and `.codex/config.toml`.
Who it is for. Testers and developers who write and fix e2e tests for web and mobile apps together with Claude Code, Codex or Cursor.
Good fit when
- The agent writes a test by walking the flow in the live app instead of guessing locators.
- A failing test is investigated: the agent runs it, stops on the failure and looks at the screen.
- After an interface change the agent opens the page and checks the result.
- A change for a pull request is shown with a video recorded from the session.
Not a fit when
- The project does not use e2e: `run_tests` does not work without an e2e config.
- A plain runner without an agent session is enough: use `e2e run` from the command line.
- The app sits behind a login that cannot be trusted to an agent: the session sees the whole app screen.
Example request
Open the sign-in page in a live session, walk through the checkout flow, record an e2e test from it and run it.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
The server works inside an e2e project: `run_tests` needs a config, and without one the agent names an engine when opening a session, for example `@e2e-dev/web`. Engines and device providers for mobile targets are set in the config. The `--max-sessions` flag caps concurrent sessions, and a second `run_tests` during a run returns `RUNNER_BUSY`. Start the client from the project directory, otherwise `npx` will not pick the installed e2e version. A test with an agent step without a recording needs a model.
How to disable. Remove the `e2e` entry from `.mcp.json`, `.cursor/mcp.json` or `.codex/config.toml`. In Claude Code use `claude mcp remove e2e`.
FAQ
How do I connect Claude Code?
Run `claude mcp add --scope project e2e -- npx e2e mcp`. For pnpm use `pnpm exec e2e mcp`, for Bun `bunx e2e mcp`.
How do I connect Codex?
Run `e2e init --yes --skills none --mcp .codex/config.toml` or add an `[mcp_servers.e2e]` entry with the command `npx` and the arguments `e2e` and `mcp`.
Is a model required?
The server needs no model. Only a test with an agent step that has no recording needs one.
Can I work without a config?
For a session yes: in a directory with no config the agent names the engine in `open_session`. `run_tests` needs a config.
Which tools does the server have?
`open_session`, `tools`, `call`, `close_session` and `run_tests`. Screen actions live in the catalog: `tools` reads it and `call` invokes them.
Discussion
No questions or reviews yet.
Related
MCP serversCLI, MCP server and typed Node.js API so an agent can verify its changes in a live mobile app on iOS, Android and more
MCP server, agent and skills for end-to-end UI tests where the agent drives the computer through vision on cloud machines
An MCP server that checks whether the agent is really done: drives the real app and returns pass, fail or couldn't tell with the file and line
kogiQA MCP gives an agent a browser that finds page elements without selectors, for debugging web apps and writing e2e tests
Electron debugging MCP server: 111 tools let an agent see the window, click buttons, read console and network and audit security.
METR
METR (MCP 1C:Enterprise Test Runner)
An MCP server that lets an agent update a 1C infobase, run YaXUnit tests and check syntax
README in short
The documentation shows how an agent works with a session, installs the server in three clients and works without a config. The reference lists tools, flags and error codes.