MCP Servers · Product Profile v2
Playwright MCP
Microsoft’s MCP server for browser automation through structured accessibility snapshots.
Evidence Tested
What is it?
Microsoft’s MCP server for browser automation through structured accessibility snapshots.
| Provider / organization | Microsoft |
|---|---|
| Homepage / project | https://github.com/microsoft/playwright-mcp |
| Repository | https://github.com/microsoft/playwright-mcp |
| MCP endpoint | UNKNOWN |
| Authentication | Target-site authentication may be configured; server itself does not establish a universal auth requirement |
| Transport | stdio and streamable-http |
| Tool count | UNKNOWN |
| Last verified | 2026-09-12 |
What evidence do we have?
Evidence state: EVIDENCE_TESTED
- Evidence level
- S1
- Frozen cases
- 10
- Valid executions
- 10
- Correct selections
- 9
- Accuracy
- 90.0%
- Wilson 95% CI
- 59.6%–98.2%
- Measurement date
- 2026-09-13
- Provider failures
- 0
Preserved miss: PW-006: browser_snapshot selected for a browser_fill_form intent.
V28 Batch 1 S1 selection-tested evidence: 9/10 valid selections (90.0%), Wilson 95% CI 59.6%–98.2%; one preserved miss. This is not S2 external gold and not a current endpoint-health claim.
What can an agent use it for?
- Navigate web pages through structured accessibility information.
- Interact with page elements and browser state in an agent workflow.
What type of tool surface does it expose?
Browser navigation, inspection, and interaction operations with multiple adjacent actions.
What remains untested or unknown?
- Current tool count depends on version and configuration.
- No public hosted endpoint is claimed; local configuration is documented.
Source provenance
Has Agent Trust Lab tested it?
Yes — existing evidence is linked above.