Puppeteer MCP Server
MCPBrowser automation for web scraping, screenshots, and page interaction through headless Chrome.
Dimension scores
Compatibility
| Framework | Status | Notes |
|---|---|---|
| Claude Code | ~ | Requires headless mode configuration, Console logs not automatically surfaced, Binary screenshot handling requires base64 flag |
| OpenAI Agents SDK | ~ | Screenshot format needs conversion for OpenAI vision API, No SSE transport limits deployment options, Complex launchOptions parameter schema translation, Evaluate tool arbitrary return types need serialization |
| LangChain | ~ | Stateful browser instance conflicts with LangChain execution model, Binary screenshot data needs custom output parser, launchOptions too complex for StructuredTool input validation, Resource reading pattern not natively supported |
Reliability
Success rate
87%
Calls made
100
Avg latency
1247ms
P95 latency
3456ms
Failure modes
- • Browser initialization timeout on rapid concurrent calls (5 instances)
- • Selector timeout errors for non-existent elements (3 instances)
- • JavaScript execution errors with malformed scripts (2 instances)
- • Type validation failures with unclear error messages (2 instances)
- • Screenshot failures with invalid dimensions (1 instance)
Code health
License
MIT
Has tests
No
Has CI
No
Dependencies
4
This is a TypeScript MCP server with good documentation and type safety. The package is published to npm (@modelcontextprotocol/server-puppeteer v0.6.2) and has a clear MIT license. The README is comprehensive (7KB) with detailed tool descriptions. TypeScript configuration is present and extends a parent config. However, several critical gaps exist: no test files or CI configuration present, README appears incomplete (cuts off mid-sentence about Docker), no CHANGELOG, and the repository structure suggests this is part of a monorepo but we cannot assess maintenance activity, issue tracking, or dependency health without access to git history and the parent repository. The dependency count is minimal (2 prod, 2 dev) which is positive. Main concerns are lack of testing infrastructure and incomplete documentation. Given the unknowns about maintenance activity and the absence of tests/CI, a score of 5 reflects a functional tool with good basics but significant gaps in quality assurance.