Perplexity MCP server
Perplexity's official Sonar MCP server: give your agent live web search, conversational answers, deep research, and reasoning.
The Perplexity MCP server is Perplexity's official integration that puts the Sonar API behind a coding agent's tool calls. Where a base model is frozen at its training cutoff, Perplexity's tools ground every answer in a live web search and return cited, present-tense results, so the agent can reason about today's documentation, release notes, prices, or news instead of guessing. It exposes four complementary tools: a raw ranked search, a conversational ask backed by sonar-pro, a deep multi-step research mode backed by sonar-deep-research, and an advanced reasoning mode backed by sonar-reasoning-pro.
The server runs locally over stdio via npx (@perplexity-ai/mcp-server) and authenticates with a PERPLEXITY_API_KEY obtained from the Perplexity API dashboard. It can also be deployed as an HTTP service (default port 8080) with Node.js or Docker when you want a shared endpoint for a team or an agent fleet. Optional environment variables tune the request timeout, the API base URL, the log level, and an outbound proxy for corporate networks. Because each tool is a thin, well-named wrapper over a Sonar model, it slots cleanly alongside docs and version-control servers without flooding the context window.
Quick install
Copy-paste configs are provided for all 8 supported clients. Pick your client below.
Add to ~/.claude.json
{
"mcpServers": {
"perplexity": {
"command": "npx",
"args": [
"-y",
"@perplexity-ai/mcp-server"
],
"env": {
"PERPLEXITY_API_KEY": "<PERPLEXITY_API_KEY>",
"PERPLEXITY_BASE_URL": "<PERPLEXITY_BASE_URL>",
"PERPLEXITY_TIMEOUT_MS": "<PERPLEXITY_TIMEOUT_MS>"
}
}
}
}claude mcp add perplexity -- npx -y @perplexity-ai/mcp-serverAvailable tools
| Tool | Description |
|---|---|
| perplexity_search | Direct web search via the Perplexity Search API, returning ranked results with metadata. |
| perplexity_ask | General-purpose conversational answers with real-time web search, backed by the sonar-pro model. |
| perplexity_research | Deep, comprehensive multi-step research backed by the sonar-deep-research model. |
| perplexity_reason | Advanced reasoning and problem-solving backed by the sonar-reasoning-pro model. |
Required configuration
- PERPLEXITY_API_KEYRequired
Perplexity API key from the dashboard at perplexity.ai/account/api. Required.
- PERPLEXITY_BASE_URLOptional
Override the Sonar API base URL. Optional; defaults to https://api.perplexity.ai.
- PERPLEXITY_TIMEOUT_MSOptional
Per-request timeout in milliseconds. Optional; defaults to 5 minutes.
- PERPLEXITY_LOG_LEVELOptional
Log verbosity: DEBUG, INFO, WARN, or ERROR. Optional.
- PERPLEXITY_PROXYOptional
Outbound HTTP proxy for corporate networks. Optional.
What you can do with it
Ground answers in the live web
When a question depends on current facts the model can't have memorized, the agent calls perplexity_ask or perplexity_search to fetch cited, up-to-date results instead of hallucinating from stale training data.
Run deep research inside the agent loop
Hand off a broad question to perplexity_research and let sonar-deep-research compile a multi-source synthesis, then continue working with the findings without leaving the editor or chat.
FAQ
- Is it free?
- The MCP server itself is open source and free to run, but it calls the Perplexity Sonar API, which is paid and billed per request against your PERPLEXITY_API_KEY. Pricing depends on the Sonar model each tool uses.
- Does it support remote/OAuth?
- The canonical mode is local stdio with an API key. There is no OAuth flow; you can self-host the server over HTTP (default port 8080) with Node.js or Docker for a shared endpoint, but it still authenticates with the Perplexity API key.