---
title: "Perplexity API MCP: review, setup & limitations · Undominated.ai"
canonical: https://undominated.ai/mcp-servers/perplexity/
description: "Official search, answer and research tools backed by Perplexity’s Search and Agent APIs, with local and hosted connection options."
---

# Perplexity API MCP: review, setup & limitations · Undominated.ai

> Official search, answer and research tools backed by Perplexity’s Search and Agent APIs, with local and hosted connection options.

[← Explore all mcp servers](/mcp-servers/)

WEB SEARCH AND RESEARCH / Perplexity

# Perplexity API MCP

Official search, answer and research tools backed by Perplexity’s Search and Agent APIs, with local and hosted connection options.

 [See setup guidance ↓](#setup)[Original source ↗](https://github.com/perplexityai/modelcontextprotocol)

SOURCE REVIEW

 Reviewed 2026-10-13
 Evidence 5 linked sources
 Publisher Perplexity
 Licence [MIT ↗](https://github.com/perplexityai/modelcontextprotocol/blob/c58e4ad254608952606f09a40934ab6cca65bfad/LICENSE)
 Revision c58e4ad25460
 [Read what was—and wasn’t—checked ↓](#review)

“The official MCP server implementation for the Perplexity API Platform”

 [Perplexity · upstream description ↗](https://github.com/perplexityai/modelcontextprotocol/blob/c58e4ad254608952606f09a40934ab6cca65bfad/README.md) Our analysis follows below.

01 / THE REASONING

## Why this made the selection.

 - Separates direct search results from synthesized answers and longer research requests.
- The current source maps answer/reason/research tools to managed Agent API presets, avoiding a stale assumption that every tool uses the older Sonar path.

### A good fit for

 - Finding current sources and then checking the original pages
- Running an explicitly scoped research request from an existing MCP client

### Weigh up before choosing

 - Synthesized answers are model output, not verified facts. Inspect citations and primary evidence before using them.
- The research tool does not expose the same domain/recency filter inputs as every other tool. Do not assume one filter applies across the server.
- Local mode still sends requests and prompt content to Perplexity. Longer research requests consume API resources, and cancellation is best-effort.

02 / THE REVIEW RECORD

## What we actually inspected.

Source review has boundaries. A clear record is more useful than a “safe” badge.

### Material inspected

 - README.md, package.json and server.json at c58e4ad254608952606f09a40934ab6cca65bfad
- Full src/server.ts tool schemas and request paths at the same commit
- LICENSE at the same commit

### Our findings

 - The direct search path and Agent API preset dispatch were read from the implementation.
- Filters, streaming and cancellation were reviewed statically; no API call, citation accuracy check or billing test was performed.
- server.ts registers four tools: perplexity_search (Search API), perplexity_ask (Agent API fast preset), perplexity_reason (medium) and perplexity_research (high). perplexity_research takes only messages; ask and reason also take recency, domain and search-context filters.
- New since c73c8561 (version 1.3.0): perplexity_search accepts search_type "fast", which the README describes as lower latency and cost.
- Self-hosted HTTP mode binds to 127.0.0.1 by default, with CORS origins empty unless ALLOWED_ORIGINS is set.
- configExample is the README's mcpServers block for other MCP clients at c58e4ad2; your_key_here is the README's placeholder.

### Not established by this review

 - No MCP server or upstream package was installed or run.
- No authenticated account, API request, client handshake or write operation was tested.

The review applies to the material and revision named here. A newer upstream release can change its behavior.

03 / PUT IT TO WORK

## Connect a server deliberately.

[Upstream setup instructions ↗](https://github.com/perplexityai/modelcontextprotocol/blob/c58e4ad254608952606f09a40934ab6cca65bfad/README.md)

DOCUMENTED COMMAND

 npx -y @perplexity-ai/mcp-server Copy command ↗

Copying does not execute this command. It may retrieve a newer version than the reviewed source.

 - For local stdio, supply PERPLEXITY_API_KEY through the client’s secret/environment settings and launch the documented package.
- Alternatively connect to https://api.perplexity.ai/mcp using the supported authenticated HTTP setup.
- Use your client’s actual configuration schema; do not paste another client’s wrapper or store API keys in shared project files.
- The publisher documents npx -yq for strict stdio clients if npm chatter disrupts initialization; use client logs to verify a clean handshake.
- The README has no standalone line for the command above: it is the command and args of the README's mcpServers block, and the tail of its claude mcp add and codex mcp add lines. It was not run.

### Before you start

 - An active Perplexity API account/key and applicable API access
- Node.js/npm for local mode, or a client supporting the hosted transport
- Network access to Perplexity APIs

### Documented configuration example

Use the client and authentication context described in the setup guidance. Configuration formats vary between hosts.

 {
 "mcpServers": {
 "perplexity": {
 "command": "npx",
 "args": [
 "-y",
 "@perplexity-ai/mcp-server"
 ],
 "env": {
 "PERPLEXITY_API_KEY": "your_key_here"
 }
 }
 }
} Copy configuration ↗

### Compatibility

Claude Code · Codex · Cursor · VS Code · Kiro

### Implementation

TypeScript

### Transports

stdio · Streamable HTTP

### API key; local or hosted transport

 - Send queries and message content to Perplexity
- Read returned search material or generated research output
- Use API-key permissions and consume provider API resources

### Cost model

MIT-licensed local implementation; Perplexity API billing/access and client/model usage are separate. No consumer-plan allowance is assumed.

04 / FOLLOW THE EVIDENCE

## The source trail.

Our notes are separate from the original resource. Check upstream before adopting a new version.

 - [Publisher README at reviewed revision ↗](https://github.com/perplexityai/modelcontextprotocol/blob/c58e4ad254608952606f09a40934ab6cca65bfad/README.md) Checked 2026-10-13 https://github.com/perplexityai/modelcontextprotocol/blob/c58e4ad254608952606f09a40934ab6cca65bfad/README.md Supports: summary, upstreamDescription, whySelected, bestFor, install, compatibility, limitations, access, configExample
- [Source licence ↗](https://github.com/perplexityai/modelcontextprotocol/blob/c58e4ad254608952606f09a40934ab6cca65bfad/LICENSE) Checked 2026-10-13 https://github.com/perplexityai/modelcontextprotocol/blob/c58e4ad254608952606f09a40934ab6cca65bfad/LICENSE Supports: license, access.cost
- [Reviewed implementation: src/server.ts ↗](https://github.com/perplexityai/modelcontextprotocol/blob/c58e4ad254608952606f09a40934ab6cca65bfad/src/server.ts) Checked 2026-10-13 https://github.com/perplexityai/modelcontextprotocol/blob/c58e4ad254608952606f09a40934ab6cca65bfad/src/server.ts Supports: review, limitations, access, transports
- [Reviewed implementation: package.json ↗](https://github.com/perplexityai/modelcontextprotocol/blob/c58e4ad254608952606f09a40934ab6cca65bfad/package.json) Checked 2026-10-13 https://github.com/perplexityai/modelcontextprotocol/blob/c58e4ad254608952606f09a40934ab6cca65bfad/package.json Supports: review, limitations, access, transports
- [Reviewed implementation: server.json ↗](https://github.com/perplexityai/modelcontextprotocol/blob/c58e4ad254608952606f09a40934ab6cca65bfad/server.json) Checked 2026-10-13 https://github.com/perplexityai/modelcontextprotocol/blob/c58e4ad254608952606f09a40934ab6cca65bfad/server.json Supports: review, limitations, access, transports

KEEP COMPARING

## Other approaches to consider.

Related by category or shared topics. These are alternatives to inspect, not a measured quality order.

 [### Zotero MCP ↗ Reads the caller’s own Zotero library. It is a community server, not a Zotero product.](/mcp-servers/zotero-mcp/)[### Algolia MCP Server ↗ Algolia's managed MCP offerings: Public MCP exposes application-scoped search and recommendations on selected indices for customer-facing agents; Productivity MCP gives user-scoped, read-only exploration and analytics; DevEx MCP serves version-matched API client snippets to coding agents without touching data.](/mcp-servers/algolia/)[### Amazon Q Index MCP ↗ ISV-only Amazon Q index search that waits for a pasted OIDC code, then assumes a supplied role before SearchRelevantContent.](/mcp-servers/awslabs-amazon-qindex/)

 [AI Tools ↗](/tools/)[Skills ↗](/skills/)[Agents ↗](/agents/)[MCP Servers ↗](/mcp-servers/)[Workflows ↗](/workflows/)

## Continue your investigation

 - [Inspect resource evidence](/tools/)
- [Choose a host](/agents/)
- [Inspect reusable workflows](/skills/)
- [Use Undominated data](/api/)
