Official (Vendor)Active· 1mo★ 4.3by Perplexity

Perplexity Ask

Perplexity Sonar models inside your agent. Real-time web answers with citations, faster than DIY RAG.

View repo Homepage

Safety & Trust

Delv Safety Grade: A

Score 85/100 · assessed 2026-04-28

Maintainer95

Permissions85

Supply chain90

Transparency75

Incidents100

Perplexity's official MCP server provides web search through their Sonar models with a clean security profile. The maintainer is a well-funded, established AI company with strong operational track record. The server requires only an API key and makes outbound network calls to Perplexity's endpoints, keeping the attack surface narrow. Distribution via npm with standard npx invocation follows best practices. Permissions are appropriately scoped to network requests and environment variable reading for the API key. The main limitation is documentation depth: while the GitHub repo and official docs exist, they're relatively sparse compared to more mature integrations. No security incidents are known. The trust model is straightforward: you're delegating web search to Perplexity's infrastructure rather than exposing local resources. For teams already comfortable with Perplexity's service, this is a low-risk addition.

Lethal Trifecta (prompt-injection exposure)

TWO OF THREE

Private dataNo

Reads secrets, credentials, private files

Untrusted inputYes

Ingests web pages, PRs, issues, emails

External commsYes

Can send data outbound

Same shape. Synthesised answers reduce the visible-attack surface but the underlying input remains untrusted.

Green flags

Official vendor package from well-funded AI company (Perplexity)
Distributed via npm with standard npx installation
Read-only search API with no local filesystem or shell access
No known security incidents or CVEs
Narrow permission scope limited to API calls and env reading

Red flags

API key stored in environment grants full account access
Documentation is thin compared to more established MCP servers
Network calls to third-party service create dependency on Perplexity uptime

Permissions requested

Outbound networkAccess secretsExternal LLM call

Assessed by Delv Editorial using public metadata. Grades are advisory and update as the ecosystem changes. They do not replace your own review of permissions and code before granting an agent access to sensitive systems.

Install

npx -y @perplexity-ai/mcp-server

Env vars needed: PERPLEXITY_API_KEY

Review

Perplexity Ask plugs the Sonar models directly into your Claude workflow, giving you web search with citations without the overhead of building your own RAG pipeline. Instead of switching tabs or prompting Claude to "search the web" and hoping for the best, you get real-time answers grounded in current sources. The value proposition is speed: you skip the dance of scraping, chunking, embedding, and retrieval that a DIY setup demands. I'd reach for this when I need factual grounding on recent events, technical documentation that changes frequently, or anything where Claude's training cutoff becomes a liability. Ask it about a library's latest API changes, current market data, or breaking news, and you get structured answers with source URLs. The citations matter because you can verify claims without second-guessing whether the model hallucinated a changelog. The quirks are mostly about scope. This isn't a general-purpose search engine replacement. It's optimised for question-answering, not exploratory browsing. If you need ten different perspectives on a topic, you'll still open a browser. The API key requirement is straightforward but adds another credential to manage. Perplexity's rate limits and pricing apply, so heavy usage will cost you beyond the free tier. Who shouldn't bother: if your work lives entirely in static codebases or historical data, this won't earn its place in your config. If you rarely need web context during agent conversations, the setup overhead isn't worth it. But for anyone doing research, technical writing, or support work where current information matters, this is faster than the alternatives. It's also officially maintained, which means fewer surprises when Claude updates. One specific workflow: I use this when debugging third-party API integrations. Instead of context-switching to read docs, I ask Claude via Perplexity Ask for the current authentication flow or rate limit details, get an answer with a link to the official docs, and keep coding. It's not magic, but it's faster than the old way.

Verdict

Install this if you regularly need Claude to answer questions grounded in current web data. Skip it if your work rarely touches anything outside Claude's training window or if you're already happy with manual search. It's a time-saver for research-heavy workflows, not a must-have for everyone.

Good at

Real-time web answers with source citations, no RAG pipeline to maintain.
Officially maintained by Perplexity, so updates track their API changes.
Faster than switching contexts to search manually during agent conversations.
Works across multiple Claude hosts, not just Desktop.

Watch out

Requires a Perplexity API key and usage counts against your rate limits and billing.
Optimised for Q&A, not exploratory search or browsing multiple perspectives.
Manual config needed for hosts beyond Claude Desktop, no one-click setup everywhere.

Getting started

1. Grab a Perplexity API key from their developer portal and set it as PERPLEXITY_API_KEY in your environment. 2. Run `npx -y server-perplexity-ask` to install the MCP server. 3. Add the server to your Claude Desktop config under mcpServers with the command pointing to the installed binary and env vars passed through. 4. Restart Claude Desktop and check the MCP icon shows Perplexity Ask as connected. 5. Test it by asking Claude a question about something recent, like "What changed in the latest Python 3.13 release?" and verify you get citations.

Works with

Claude DesktopClaude CodeCursorWindsurfClineZed