Perplexity MCP Server
Not yet source-checkedOfficial MCP server for Perplexity's Sonar search, research, and reasoning APIs
Machine-drafted from the vendor's site · not yet source-checkedProfile updated Link checked Pricing captured How we rate
In short
The Perplexity MCP Server is the official integration that allows MCP-capable clients to access Perplexity's Sonar search, research, and reasoning APIs. It is best for developers and agent builders who need first-party, cited web grounding and current-events awareness without building their own search infrastructure.
Developers and agent builders who want first-party, cited web search and deep research inside Claude Desktop, Cursor, VS Code, or a custom MCP client without wiring their own search + LLM stack.
Teams that need offline or air-gapped retrieval, want to search a private corpus, or are cost-sensitive about agents that call research tools in loops.
Perplexity MCP is the official Model Context Protocol server maintained by Perplexity (ppl-ai/modelcontextprotocol on GitHub) that lets any MCP-capable client hand off live web search, reasoning, and research to Perplexity's Sonar-family Agent and Search APIs. It exposes four tools: perplexity_search returns ranked web results with metadata plus recency and domain filters; perplexity_ask runs a conversational query against the fast Sonar preset with real-time browsing; perplexity_research kicks off a long-running deep-research preset that can spend minutes gathering and cross-checking sources; and perplexity_reason drives the medium reasoning preset for multi-step analysis. Presets abstract model choice, retrieval setup, and compute budgets, so callers pick an intent instead of a model string. The project ships two deployment modes: a hosted remote endpoint at https://api.perplexity.ai/mcp that any client with an API key can attach to in one line, and a local npm package (@perplexity-ai/mcp-server) that runs over stdio or HTTP for Cursor, VS Code, Claude Desktop, Claude Code, Windsurf, and self-hosted Docker setups. It is aimed at developers and agent builders who want first-party grounding, citations, and current-events awareness inside their assistants without wiring a search stack themselves. Typical workflows include coding agents pulling up-to-date library docs before writing code, research assistants producing cited briefs, customer-support bots looking up current policies, and Claude Desktop power users adding a 'search the web and cite it' capability that isn't fabricating URLs. Because it's the vendor's own server, tool schemas track new Sonar features (recency filters, structured citations, preset changes) as they ship, and legacy model IDs like sonar-pro have already been removed in favor of presets.
If you're already paying for Perplexity's API, this is the obvious way to plug it into an MCP client — official, MIT-licensed, and the hosted endpoint means one line of config. The catch is the deep-research preset: it's genuinely useful but slow and pricey, so gate it behind explicit user intent rather than letting an autonomous agent fire it in a loop.
— The AI Tool Bible editorial team
Pros
- ✅ Official first-party server, so tool contracts and citations stay in sync with Perplexity's API
- ✅ Four distinct tools cover raw search, fast ask, deep research, and reasoning without changing endpoints
- ✅ Remote hosted endpoint (api.perplexity.ai/mcp) means zero-install setup for most MCP clients
- ✅ Returns structured citations and source URLs, which agents can surface or verify
- ✅ Presets hide model churn — callers pick intent, not a model string that may be deprecated
- ✅ MIT-licensed reference implementation you can fork, self-host, or embed in a Docker image
- ✅ Recency and domain filters on perplexity_search let you scope agents to fresh or trusted sources
Cons
- ⚠️ API usage is metered and can get expensive fast if agents call perplexity_research in a loop
- ⚠️ Deep-research preset can take several minutes per call, which stresses MCP client timeouts
- ⚠️ Legacy model names were removed, so older client configs pointing at sonar-pro/sonar-reasoning-pro break
- ⚠️ Quality depends entirely on Perplexity's retrieval — you can't swap in a private index or your own crawler
- ⚠️ Rate limits and cost visibility live in Perplexity's dashboard, not the MCP surface
- ⚠️ Only four tools exposed; more niche Perplexity API features (async jobs, spaces) aren't wrapped yet
Pricing (captured )
| Tier | Monthly | Per seat | Key features |
|---|---|---|---|
| Free | $0 USD per month | — |
|
| Team | $4 USD per user/month | Yes |
|
| Enterprise | Starting at $21 USD per user/month | Yes |
|
Prices change; we show the grid as captured on Oct 8, 2026. Check current pricing on the vendor's site →
Facts
- API available
- Yes
- Free trial
- Yes
- Open source
- Yes (license: MIT)
- GitHub stars
- 2,595 (checked Oct 29, 2026)
- Last GitHub push
- Oct 21, 2026
- First commit
- Mar 2025
- Model used
- Perplexity Sonar family (via Agent and Search API presets)
Where this tool sits
Use cases
Frequently asked
- Which MCP clients are supported?
- The server supports any MCP-capable client. It includes a local npm package compatible with Cursor, VS Code, Claude Desktop, Claude Code, and Windsurf, as well as a hosted remote endpoint for general use.
- What specific tools does the server provide?
- It exposes four tools: perplexity_search for ranked web results, perplexity_ask for conversational queries with real-time browsing, perplexity_research for long-form deep research, and perplexity_reason for multi-step analysis.
- How is the pricing structured?
- Pricing is paid, with a free tier at $0 USD, a Team plan at $4 USD per user per month, and an Enterprise plan at $21 USD per user per month.
- Can I self-host this server?
- Yes, the project is MIT-licensed and ships as a local npm package that can run over stdio or HTTP. It can be embedded in Docker images for self-hosted setups.