Perplexity MCP Server – Search and Research
Perplexity's official MCP server connects AI assistants to the Perplexity API Platform. Use it when an agent needs real-time web search, current answers, deep research, or advanced reasoning powered by Perplexity Sonar models and the Search API.
Overview
Perplexity's official MCP server gives AI assistants access to Perplexity's
real-time search, research, and reasoning capabilities. It is useful when an
agent needs current web information or a research-oriented answer rather than
relying only on static model training data.
What the MCP server enables
The official server exposes four core tools:
perplexity_searchfor direct web search using the Perplexity Search API,
returning ranked search results with metadata.perplexity_askfor general conversational answers with real-time web search
using thesonar-promodel.perplexity_researchfor deep, comprehensive research using the
sonar-deep-researchmodel.perplexity_reasonfor advanced reasoning and analytical problem solving
using thesonar-reasoning-promodel.
The perplexity_reason and perplexity_research tools also support the
optional strip_thinking parameter, which can remove thinking tags from
responses to save context tokens in clients that do not need that output.
When to use it
Use Perplexity MCP when an AI workflow needs fresh information, source-aware
exploration, market or technology research, comparative analysis, or a
reasoning-focused response. Practical examples include researching recent
software releases, collecting sources on a business topic, checking current
facts before writing code or strategy, preparing a detailed report, or using
web search inside an IDE or desktop AI client.
Connection and authentication
Local stdio setup uses npx -y @perplexity-ai/mcp-server and requires
PERPLEXITY_API_KEY. Optional environment variables include
PERPLEXITY_TIMEOUT_MS for long-running requests, PERPLEXITY_BASE_URL for a
custom API base URL, PERPLEXITY_LOG_LEVEL for logging, and
PERPLEXITY_PROXY, HTTPS_PROXY, or HTTP_PROXY for corporate proxy
environments.
For shared or cloud deployments, the official repository documents HTTP server
mode. The HTTP server can be started from source with npm run start:http after
installation and build steps, or deployed with Docker. By default, it listens on
port 8080 and exposes MCP at http://localhost:8080/mcp.
Key considerations
Perplexity MCP calls use the Perplexity API Platform and require a valid API key.
Very long research tasks may need a higher timeout setting. Corporate networks
may require explicit proxy configuration, and the server checks proxy variables
in the documented order: PERPLEXITY_PROXY, then HTTPS_PROXY, then
HTTP_PROXY. Some strict MCP clients can fail when npx writes installation
messages to stdout; the official troubleshooting guidance suggests using
npx -yq instead of npx -y to suppress that output. Review outputs before
taking business, legal, medical, financial, or deployment actions, because web
search and model-generated analysis still require human validation.
Supported Transports
stdio
Command: npx
Args:
-y@perplexity-ai/mcp-server
streamable_http
URL: http://localhost:8080/mcp
Frequently Asked Questions
- When should an AI agent use the Perplexity MCP server?
- Use it when a workflow needs current web information, ranked search results, source-aware answers, deep research, or advanced reasoning, such as researching recent releases, comparing vendors, collecting sources, or preparing a report from live web context.
- What does the Perplexity MCP server add to an AI agent's capabilities?
- It gives the agent access to Perplexity's Search API and Sonar-powered tools for real-time web search, conversational answers, deep research, and reasoning, rather than relying only on static model knowledge.
- What can an AI agent access or manage through Perplexity MCP?
- The agent can call `perplexity_search`, `perplexity_ask`, `perplexity_research`, and `perplexity_reason`. These tools retrieve and analyze web information; they do not manage user files, accounts, or external business systems.
- How is authentication configured for the Perplexity MCP server?
- The MCP server requires PERPLEXITY_API_KEY from the Perplexity API Portal. Store the key in an environment variable or secret manager. Optional settings include timeout, base URL, log level, and proxy environment variables.
- Which transport should be used for the Perplexity MCP server?
- Use local stdio with `npx -y @perplexity-ai/mcp-server` for desktop and IDE clients. Use local or hosted Streamable HTTP at `/mcp` only when running the official HTTP server mode for shared, cloud, or multi-client deployments.