Run a research agent

Run a research agent

Starts an agent that searches, navigates and reads the web to return the data a prompt describes, as JSON.

Use it when the source pages are not known, such as a company plus the fields you need about it; pass a JSON schema for structured output and optional seed URLs. The agent runs as a job: the API returns an id to poll with GET /agent/{jobId}, and the MCP server pairs firecrawl_agent with firecrawl_agent_status. Firecrawl labels Agent a research preview.

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.

{ "mcpServers": { "firecrawl": { "url": "https://mcp.firecrawl.dev/v2/mcp-oauth" } } }

Call the MCP tool firecrawl_agent.

Server URL: https://mcp.firecrawl.dev/v2/mcp-oauth

CLI (official)

Install the command, then confirm it runs.

npm install -g firecrawl-cli
firecrawl --version

Run firecrawl agent.

Set $FIRECRAWL_API_KEY in your environment first (get a key: https://www.firecrawl.dev/app/api-keys).

API (official)

• Base URL: https://api.firecrawl.dev/v2
• Endpoint: POST /agent
• Auth: send the header Authorization: Bearer $FIRECRAWL_API_KEY
• Get a key: https://www.firecrawl.dev/app/api-keys
• Docs: https://docs.firecrawl.dev/api-reference/introduction

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.