Search and docs lookup you would do anyway; `check_deprecated_code` runs on code you already have and tells you what has moved on since it was written — the class of problem nobody goes looking for. Costs land on your Perplexity key rather than here, and the built-in backoff is what keeps a long session from hitting the rate limit.
A research assistant backed by Perplexity's Sonar models. Four tools cover the questions that interrupt coding: general search, documentation lookup for a library, finding candidate APIs for a job, and checking whether the code in front of you uses something deprecated. Conversation history is kept in SQLite so follow-up questions stay coherent.
- `search` answers a general question, with brief, normal and detailed levels of depth
- `get_documentation` returns usage and examples for a named library or API, including common pitfalls
- `find_apis` proposes APIs for a task and compares them on features, pricing and integration effort
- `check_deprecated_code` reads code for outdated patterns and gives migration guidance
- `get_request_status` reports on an in-flight request
- Adaptive rate limiting with exponential backoff, plus connection pooling and automatic retries, so a burst of questions does not trip the API
Node.js and a Perplexity API key from `https://www.perplexity.ai/settings/api`. Published as research-mcp version 1.0.0; the documented path is clone, install dependencies, `npm run build`, then point your client's `command` at node and its `args` at `build/index.js`. `PERPLEXITY_API_KEY` is required; `PERPLEXITY_MODEL` selects the model and the sample config uses `sonar-reasoning`. Because every tool is a read, all five can safely be listed under `autoApprove`.
Build from source — clone the repository and build it, then point your client at the binary