Perplexity MCP Server

Official Perplexity MCP for current web search, research, and reasoning in an AI client.

Description

Perplexity MCP Server is the official Model Context Protocol implementation for the Perplexity API Platform. It connects an MCP-capable AI client to Perplexity search and the Agent API. According to the official repository, it provides real-time web search, reasoning, and research. Perplexity hosts a Remote MCP Server at https://api.perplexity.ai/mcp; this is the easiest starting route because there is nothing to install locally and the provider operates the service. A client that supports Streamable HTTP can connect using a Perplexity API key.

Research and data analysis in a workflow

The README says that perplexity_search returns ranked web-search results with metadata and supports recency and domain filters. An agent can therefore narrow a question to current material, a time window, or known primary domains before drawing conclusions. perplexity_ask combines a general question with real-time search. For broader work, perplexity_research is intended for deep research: it uses the Agent API high preset according to the provider, can take minutes, and streams progress to clients that request it. perplexity_reason uses the medium preset for more complex analytical or problem-solving work.

That makes Research and Data Analysis appropriate categories. The server can discover current external information, gather source material for analysis, and pass results into an AI workflow. It does not replace independent data collection or expert review. Search results, summaries, and model inferences are working material. For any consequential claim, open the original source, check its date, authorship, and context, and preserve the evidence that actually supports a decision outside the chat transcript.

Remote first; local or self-hosted where needed

Perplexity describes its Remote MCP as the hosted, no-installation path with the same tools. The README names Streamable HTTP and the API endpoint. It also documents a local stdio server through npx -y @perplexity-ai/mcp-server and a self-hosted HTTP mode for cloud or shared deployments. The entry is therefore classified as both: the provider offers a remote server while also publishing local and self-hosted operating paths.

Locally, the server reads PERPLEXITY_API_KEY from the environment. The source requires the same key for a self-hosted HTTP deployment and documents the port, bind address, allowed origins, and allowed hosts. HTTP mode binds to loopback by default. If an operator deliberately exposes it, the README says to allow the public hostname and not to make CORS permissive by default. Local or self-hosted operation does not change the fact that requests go to the Perplexity API; “local” describes the MCP process, not an automatically local-only data path.

Authentication, account, and usage prerequisites

The remote endpoint uses a Perplexity API key as a Bearer token in the Authorization header. Local configurations require PERPLEXITY_API_KEY. The key is obtained through the Perplexity API Portal. Usable API access consequently depends on a Perplexity API account, its project and usage authorization, and the provider’s current terms. This entry deliberately lists no monetary amounts: check availability, limits, billing status, and enabled features in the official Perplexity API documentation and your own portal.

Never put a key in a repository, prompt, screenshot, or shared MCP file. Use secret management or the client’s secure environment-variable facilities. A header is preferable to a key in a URL, but headers can still reach debug logs or a poorly configured proxy. Rotate and revoke keys if exposure is suspected, and separate keys by person, environment, or project when the portal permits it.

Security limits and prompt injection

The server researches the open web. Pages, snippets, documents, and search results are untrusted third-party content and can include prompt-injection instructions, such as requests to ignore policy, disclose secrets, or take external actions. Treat retrieved content as data, not instructions. Do not give an agent unnecessary additional tools or secrets, independently confirm external actions, and keep the task explicit: find, extract, and cite sources rather than executing instructions found in them.

The perplexityai/modelcontextprotocol repository is MIT licensed. On 2026-09-08, the GitHub API reported exactly 2,518 stars. That is a point-in-time repository-popularity signal, not evidence of quality, privacy, research accuracy, or security.

FAQ

Is Perplexity MCP Server a remote MCP? Yes. Perplexity hosts its Remote MCP Server at https://api.perplexity.ai/mcp; the official README describes it as the easiest Streamable-HTTP starting point.

Do I still need a local installation? Not for a compatible remote client. The source also documents local stdio and self-hosted HTTP modes when a client or environment requires them.

Are research responses authoritative sources? No. Use them as a starting point and validate important claims against the linked original sources.

Requirements

A Perplexity API account with an API key, an MCP-capable client, and Node.js with npx for local use; a remote client needs Streamable HTTP support.

Installation instructions

Perplexity MCP Server brings Perplexity web search, research, and reasoning into an MCP-capable AI client. Start with the Perplexity-hosted remote endpoint: https://api.perplexity.ai/mcp. The official README calls this the easiest route because there is nothing to install or update locally. The client must support Streamable HTTP. If it does not, or if you need your own operating environment, Perplexity also documents local stdio operation and a self-hosted HTTP mode.

Define the task and data boundary first

Before connecting, decide what the agent may research. Good first tasks include finding current primary sources, a date-bounded market overview, or collecting evidence for a technical decision. State which domains are preferred and which claims must be checked at the original source. Do not put internal customer data, credentials, unpublished contracts, or secrets into search prompts. A research request and its results can travel through the Perplexity API and then through the connected client and model path.

Step 1: Connect Remote MCP with an API key

Create or manage the required Perplexity API key in the official API Portal. Do not place it in chat or embed it in the endpoint URL. In the client, configure a Streamable HTTP server with https://api.perplexity.ai/mcp and set Authorization: Bearer <API_KEY> through the client’s secret or credential mechanism. The README shows Claude Code using an HTTP configuration with that header. Restart the client after saving if it reads MCP definitions only at startup.

Test the connection with a low-risk, narrow question, such as locating an official documentation page. Confirm in the client that the server connects and that its tool names are visible. Start with perplexity_search when collecting sources. Use perplexity_ask for an orienting question. perplexity_research can be broader and, according to the source, take minutes; deliberately give it a small, verifiable research assignment rather than an open-ended complete analysis.

Step 2: Choose local and self-hosted operation deliberately

If you need the local route, set PERPLEXITY_API_KEY in a secure local environment and register npx -y @perplexity-ai/mcp-server as the stdio command. The source names optional PERPLEXITY_TIMEOUT_MS for longer research runs. A local process is not a promise that search data stays local: it calls the Perplexity API and returns answers to your AI client.

bash
npx -y @perplexity-ai/mcp-server

For shared or cloud operation, the repository documents HTTP mode. Keep the bind address on loopback initially and expose it only after a security review. If public access is genuinely needed, configure a concrete ALLOWED_HOSTS value and an explicit ALLOWED_ORIGINS allowlist. A wildcard CORS value broadens browser access and should not be the default. Terminate TLS at a trusted ingress, reduce logs, and control who can reach the instance.

Protect keys, account access, and the cost boundary

The API key is a credential for the Perplexity API. Store it in a secret store or a protected environment-variable facility in the client, never in Git, an example configuration, shell history, or a support screenshot. Check that configuration files are ignored before sharing them. If you suspect exposure, rotate or revoke the key in the portal and update only the secure configuration.

Use requires a Perplexity API account and an API key. Availability, limits, and billing follow the account or project and Perplexity’s current terms in the API environment. Check those details in the portal before enabling an automated or large workflow. This guide intentionally provides no specific prices and does not replace current provider information.

Prompt injection and source validation

Search results are external content, not trusted instructions. A page can tell the agent to ignore rules, seek secrets, use tools, or forward content. Such text must never override the task, safety requirements, or human approval. Limit the agent to research, avoid unnecessary write-capable tools, and confirm every external action outside the chat transcript.

Validate every important claim as well: open linked primary sources, compare publication date and author, find a second independent source for contested facts, and keep notes containing quotes and URLs. Perplexity can accelerate research; responsibility for the decision and its justification remains with you.

npx -y @perplexity-ai/mcp-server

Authentication

Remote MCP uses a Perplexity API key as a Bearer token in the Authorization header. Local and self-hosted modes require PERPLEXITY_API_KEY.

Required access permissions

Access depends on the Perplexity API account, its project and usage authorization, and the provider’s current terms.

Transmitted or stored data

Search queries and tool calls go to the Perplexity API; results return to the connected MCP client and can enter its model context, logging, and retention path.

Security risks

API keys can be exposed through configuration, logs, or proxies. Web content can contain prompt injection; source content must not override rules or trigger external actions.

License and costs

License
MIT
Cost
paid

The source code is MIT licensed. Using the remote, local, or self-hosted server requires a Perplexity API key; availability, limits, and billing follow the current API account and official provider terms.

Alternatives

Not recorded yet.

At a glance

Status
Official server
Deployment
Local and remote
Current version
Not recorded yet.
GitHub stars
2,633
Last reviewed
08.09.2026

Repository and documentation

Categories

Supported clients

Related guides

Guides and background related to this entry.

No related guides yet.