Playwright MCP and a testing MCP server connect an agent to an application, but they leave different artifacts behind.
The short answer
Playwright MCP gives an agent a browser for the session it is in. It can navigate, click, take snapshots, and generate Playwright code at the end. A testing MCP server gives an agent a test suite it can add to. It can author a test, run it, read the failure, and leave the test in your repository. If you want an agent to interact with a page today, use Playwright MCP. If you want the check to still run next month in CI, use a testing MCP server.
What each server actually exposes
Playwright MCP is Microsoft's open-source server. Its browser tools return a structured page snapshot on every call. The snapshot includes the rendered DOM, accessibility tree, and console.
Momentic's MCP server is local. You start it with npx momentic mcp. Its tools preview a candidate step against the live page, return a compressed screenshot and a short status, commit the step on success, run the saved test, and hand back the failure.
Comparison table
| Playwright MCP | Momentic MCP | |
|---|---|---|
| What the agent produces | Playwright code, materialized locators and string literals at generation time | act, assert and extract steps that describe user intent, saved as YAML |
| Where the artifact lives | your repo, as a spec file | your repo, as a YAML test file |
| Authoring loop | interact from memory, then generate code at the end | preview each step against the live page, commit it on success |
| Editing a test later | full browser and session reset, the agent replays from memory | splice individual steps, the browser session persists across edits |
| Context the agent carries | a full page snapshot per tool call, accumulating across the session | a screenshot and a short status per call, with locator resolution done server-side |
| Replay | a changed selector or a changed string is a test failure | a step replays from a multi-modal cache in milliseconds with no model call, and re-resolves in place when the page changes |
| Waiting | hard-coded waitForTimeout in the generated spec | built-in waits for navigation, load, screenshots, DOM mutations and same-origin requests |
| Running the suite in CI | whatever you build around playwright test | npx momentic run, with labels, sharding, JUnit output and a non-zero exit code on failure |
| Mobile | web only | the same YAML and the same CLI for iOS and Android |
| Best for | ad hoc browser work and teams who want standard Playwright code | a regression suite an agent maintains with you |
Set it up
The server needs MOMENTIC_API_KEY in its environment. Follow the client-specific setup in the MCP server documentation .
Claude Code
Run this command from the project that should use the server. Verify the connection with claude mcp list, then use /mcp inside Claude Code.
claude mcp add --transport stdio momentic -- npx momentic mcp --config /absolute/path/to/momentic.config.yamlCodex
Run this command to save the server configuration to ~/.codex/config.toml. Restart Codex before you run /mcp.
codex mcp add momentic -- npx momentic mcp --config /absolute/path/to/momentic.config.yamlCursor
Open the command palette, select View: Open MCP Settings, click Add a new Custom MCP server, paste this entry, and restart Cursor.
{
"mcpServers": {
"momentic": {
"type": "stdio",
"command": "npx",
"args": [
"momentic",
"mcp",
"--config",
"/absolute/path/to/momentic.config.yaml"
],
"env": { "MOMENTIC_API_KEY": "your-api-key" }
}
}
}Restart the client, start a new chat, and ask the agent which MCP tools it has. You should see momentic_* tools such as momentic_get_artifacts.
The momentic-test, momentic-spec, and momentic-maintain skills install alongside the server. See the MCP server documentation .
The documentation also covers VS Code Copilot, OpenCode, Devin, and other clients.
What the agent can do once it is connected
Your agent authors a test step by step and captures a screenshot after each step. It runs the test it just wrote, then reads back the failing step, the DOM, and the console. After a UI change it updates the test in your repository, so the change arrives in a pull request.
Playwright MCP is the right call if
- You already have a Playwright suite and want authoring help inside it.
- You want the artifact to be Playwright code your team already reviews.
- You need a browser in an agent session and have no suite yet.
Momentic's MCP server is the right call if
- You want tests to survive the next UI change without a code review.
- You want one suite across web, iOS, and Android.
- You want a merge gate rather than a session transcript.
How to try both
Read the deep technical comparison for the detail behind the table. Then follow the MCP setup guide to connect a client. Use the CLI when you are ready to run the saved suite in CI. Talk to us if you want help choosing a starting point.