---
ref: tool:perplexity/deep-research
name: Run deep research
company: company:perplexity
workflows: []
access: [mcp, api]
tags: [capability:research-accounts, category:web-search, has:api, has:mcp]
docs: https://docs.perplexity.ai/api-reference/agent-post
updated: 2026-09-27
---

# Run deep research

Researches a question across many web sources with extended reasoning and returns a thorough answer with citations.

`perplexity_research` is backed by the Agent API's `high` preset: on the API,
send `preset: "high"` with the question as `input`. A long run can go in the
background with `background: true` and be polled by its response ID.

## Set up

Use the first option your agent supports.

### MCP (official, remote)

Add this server to your agent's MCP settings, then sign in when asked.

```json
{ "mcpServers": { "perplexity": { "url": "https://api.perplexity.ai/mcp" } } }
```

Call the MCP tool `perplexity_research`.

Server URL: https://api.perplexity.ai/mcp

### API (official)

- Base URL: https://api.perplexity.ai
- Endpoint: `POST /v1/agent`
- Auth: send the header `Authorization: Bearer $PERPLEXITY_API_KEY`
- Get a key: https://console.perplexity.ai/project/keys
- Docs: https://docs.perplexity.ai/docs/getting-started/overview

Before doing anything else, make one read-only call to confirm access.

## Rules

- Ask the user before anything that sends messages, costs money, or changes data.
- Never print API keys.
