This profile is part of our MCP server directory, within The MCP Guide. Profiles are checked against each vendor's own documentation. Need a server built? See MCP development.
The Perplexity MCP server lets an AI app search the web, get quick cited answers, reason through problems and run deep research. Perplexity publishes it as "the official MCP server implementation for the Perplexity API Platform." It comes as a hosted remote server and as an open-source local package.
Checked against: Perplexity MCP server docs on 28 October 2026. eSEOspace did not install or test this server.
Key Takeaways
- Perplexity's official server offers four tools:
perplexity_search,perplexity_ask,perplexity_researchandperplexity_reason. - The hosted server lives at
https://api.perplexity.ai/mcp, with a local npm package as a fallback. - You can sign in with OAuth or send an API key as a bearer token.
- Tool calls are billed to your Perplexity API account at standard API pricing.
- All four tools read from the web; none of them change your data.
At a glance
| Field | Detail |
|---|---|
| Publisher | Perplexity |
| Official? | Yes |
| Hosting | Both: remote URL https://api.perplexity.ai/mcp and local package @perplexity-ai/mcp-server |
| Transport | Streamable HTTP (remote); stdio or self-hosted HTTP (local) |
| Sign-in | OAuth ("Sign in with Perplexity") or API key |
| Read or write | Read only (search and answers) |
| Price | No separate MCP price listed; calls billed at standard API pricing |
| Source code | github.com/perplexityai/modelcontextprotocol |
| Status | Not stated |
What it can do
The GitHub README documents four tools:
perplexity_search: direct web search through the Perplexity Search API. It returns ranked results with metadata and supports recency and domain filters.perplexity_ask: conversational answers with real-time web search. It uses the Agent APIfastpreset.perplexity_research: deep research for detailed reports. It uses thehighpreset, and runs "can take minutes."perplexity_reason: step-by-step reasoning and analysis. It uses themediumpreset.
perplexity_ask and perplexity_reason accept max_tokens and max_tokens_per_page. The README says these limit retrieved search content, not answer length.
How to connect it
For the hosted server, the README gives this Claude Code command:
claude mcp add --transport http perplexity https://api.perplexity.ai/mcp --header "Authorization: Bearer YOUR_API_KEY"The docs say you can also add the server "with no credentials." Your client then opens a browser to sign in and pick the API organization to bill.
For the local server, get an API key from the Perplexity API Portal. The README's config for clients that use the mcpServers format is:
{
"mcpServers": {
"perplexity": {
"command": "npx",
"args": ["-y", "@perplexity-ai/mcp-server"],
"env": {
"PERPLEXITY_API_KEY": "your_key_here"
}
}
}
}The README lists config file paths for Cursor, Claude Desktop, Kiro, Windsurf (now Devin Desktop) and VS Code. It also covers Codex, a Claude Code plugin install, corporate proxies and a self-hosted HTTP mode. For very long research runs, raise PERPLEXITY_TIMEOUT_MS (default 5 minutes).
Security notes
Keys. The docs say "Never commit or share API keys." They advise keeping keys in global config, not shared workspace files. Our page on MCP secret leakage explains why.
OAuth scope. Per the docs, signing in grants API usage only. The connection "can't create API keys, view balances, or manage the organization."
Self-hosting. In HTTP mode, the server binds to 127.0.0.1 by default. Setting BIND_ADDRESS to 0.0.0.0 exposes it on all interfaces, and the README pairs that with ALLOWED_HOSTS and ALLOWED_ORIGINS settings. Do not expose it to a network you do not trust.
Cost control. Because calls bill to your API account, a long agent loop can run up usage. Set limits in your Perplexity account and keep tool approvals on for perplexity_research.
Untrusted results. Search answers draw on public web pages. Treat them as untrusted input, as covered in prompt injection through MCP tool results. The MCP security center has the full checklist.
Pricing
Perplexity's docs state: "Tool calls are billed at standard API pricing to the API key you connect with." On the Perplexity API pricing page, the Search API costs $5.00 per 1,000 requests ($1.00 per 1,000 with Fast Search). The Agent API bills per token by model, plus tool fees. Prices are as of October 2026.
Alternatives
- Exa MCP server: hosted search and page fetching, with keyless free use.
- Tavily MCP server: search, extract, crawl, map and research tools.
- Brave Search MCP server: Brave's independent index, with web, news, image and local search.
More options are on our MCP server profiles hub.
Next step
If you want AI answer engines like Perplexity to cite your brand, the server is only half the story. Our MCP marketing team helps MCP and AI companies get found in AI search.
Frequently asked questions
Is the Perplexity MCP server free?
The server code is open source, but tool calls are billed. The docs say calls are charged at standard API pricing to the key or organization you connect. Check the pricing page for current rates.
Do I need an API key?
Not always. The remote server supports OAuth, where you sign in with Perplexity and pick an organization to bill. Clients without OAuth support can send an API key as a bearer token.
Which tool should I use for deep research?
perplexity_research. The README says it uses the Agent API high preset and runs can take minutes. For quick questions, perplexity_ask is the lighter option.
Can the server change anything in my Perplexity account?
The documented tools search and answer only. The docs add that an OAuth connection cannot create keys, view balances or manage the organization.
