All servers / Browser / mlola MCP server
mlola MCP server
mlola-mcp: a Model Context Protocol server that lets coding agents run browser tasks and get back what each run proves.
Tested by ConnectATDScore 86 out of 100. Tested 2026-11-19 on package version 0.18.0. Package: @mlola/browser-mcp · source
Score breakdown
| Runs | 18 / 20 | |
|---|---|---|
| Tool descriptions | 20 / 20 | |
| Package safety | 20 / 20 | |
| Maintenance | 20 / 20 | |
| Adoption | 8 / 20 |
What we saw
- Starts and answers the MCP handshake: yes (9.0 seconds including download)
- Tools offered: 5
- Last published: 2026-11-13
- Weekly downloads (npm): 732
- License: MIT · Direct dependencies: 4
Want this server set up for your team?
ConnectATD builds AI automation and wires up MCP servers like this one. Tell us what you are trying to do.
Talk to ConnectATDTools and how they are described
Agents choose tools from these descriptions. Under each name you can see how many of its parameters are explained.
browser_run7 of 7 parameters describedDrives a real browser through a task and returns its outcome: `verified` (independent evidence confirms the success criteria), `failed`, or `inconclusive` (the run ended without proof; do not treat it as success). State successCriteria whenever you can: without them most goals can only end inconclusive. Typical use: after changing a web app, check that a user journey still works. Give values to type in `inputs`, keyed by field label. Risky actions (payments, deletion, credentials) are never performed unattended.
browser_verify7 of 7 parameters describedChecks stated criteria against a web app's real state and returns `verified`, `failed` or `inconclusive`, with a receipt anyone can check offline. No model is used and nothing is clicked or typed. Use it after you (or another tool, e.g. a Playwright browser) changed or exercised a web app: `url` opens a page fresh in MLola's own browser; `attach: true` reads the browser the operator configured (observe-only), optionally the tab whose URL contains `page`. Criteria can ask the server directly (http_readback), not only the page. With `attach`, set watchSeconds and call BEFORE the work to also judge server and console errors during it; a single look after the work cannot count them and says inconclusive rather than 'no errors'. For an agent's purchase under AP2, pass the mandates (cart; intent and payment when there are) as `mandate`: the cart becomes criteria against the merchant's order record, and the verdict also needs the mandates to agree (payment total, merchant, expiry).
browser_prove3 of 3 parameters describedRuns the approved contract of a change (its acceptance criteria, from the PRD or change request) against the app and returns PROVEN, NOT PROVEN or NOT PROVEN YET, criterion by criterion, with a signed proof. Each criterion is checked at the layer its contract names: the CI test report, the API, or the page in Mlola's own browser. No model is used. Call it before saying a change is done; when a criterion fails, fix the code and call again. First run the tests so they write the JUnit report the contract names: this tool reads that report and does not run tests, and a report older than your last edit proves nothing. Do not edit the contract or the tests that prove it: the proof records that and sends it to a person. `contract` is a file name in the contracts directory the operator configured; `baseUrl` is where the app runs (e.g. your dev server); `base` (e.g. origin/main) compares with the base branch.
browser_get_run1 of 1 parameter describedReturns a past run's result and a compact timeline (steps, policy decisions, verification), to see why a run failed or ended inconclusive.
browser_list_runs1 of 1 parameter describedLists recent runs, newest first, with their status.
Tested in a sandbox with an empty environment: no API keys, a throwaway folder, nothing from a real account. A passing test shows the server runs and responds. It is not a security audit.