# Firecrawl MCP

Canonical: https://cqbase.net/en/entry/firecrawl-mcp
Chinese: https://cqbase.net/zh/entry/firecrawl-mcp
English: https://cqbase.net/en/entry/firecrawl-mcp

> The official Firecrawl MCP server for search, scrape, and page parsing.

firecrawl/firecrawl-mcp-server is Firecrawl’s official MCP server, MIT licensed, with the product at firecrawl.dev. Search, scrape, crawl, and parse are tools an agent can call when an answer needs a source instead of a remembered description of a page.

Check that you are allowed to fetch the target, and check the quota on the API key. Treat fetched text as source material. Do not republish it as your own article.

## What problem it solves

[firecrawl/firecrawl-mcp-server](https://github.com/firecrawl/firecrawl-mcp-server) is Firecrawl’s official MCP server (MIT license). It lets clients such as Claude, Cursor, and Codex call Firecrawl search, scrape, parse, crawl, and map tools. Official docs now recommend the hosted HTTP remote server first; you do not have to keep a long-running Node process on your machine.

The important split is this: the hosted endpoint can be used without an API key, but official docs say that path only exposes Search, Scrape, and Parse, and it is rate-limited. Crawl, Map, and Agent still need an API key. Keys start with `fc-` and belong in a request header. Do not put a key in the URL, and do not paste it into the chat.

## Who it is for

It is for people who already use an MCP-capable client and want a controlled way to let an assistant fetch pages. Plans shown on the Firecrawl homepage on 2026-11-15 were: Free, 1,000 credits a month, no credit card; Hobby, $19 a month or $16 on the annual price; Standard, $99 / $83; Growth, $399 / $333; Scale, $749 / $599; Enterprise, custom pricing. Credits are shared across the product and the API. Official docs describe 1 credit as roughly one page of a basic scrape or crawl; search and other operations are metered as the pricing docs describe. Prices can change; use [firecrawl.dev](https://www.firecrawl.dev/) as the source of truth. This page did not find a published fixed daily cap for the keyless hosted path, so it does not invent one.

## How to install it

Verified on 2026-11-15 (Asia/Shanghai). The commands come from the repository README and the official MCP docs. The keyless hosted endpoint is:

```text
https://mcp.firecrawl.dev/v2/mcp
```

The OAuth endpoint is `https://mcp.firecrawl.dev/v2/mcp-oauth`. That is an MCP endpoint, not a page you open in a browser. When you use an API key, you still use `/v2/mcp` and add `Authorization: Bearer <FIRECRAWL_API_KEY>`.

Add the keyless remote server in Codex with:

```bash
codex mcp add firecrawl --url https://mcp.firecrawl.dev/v2/mcp
```

In Claude Code:

```bash
claude mcp add --transport http firecrawl https://mcp.firecrawl.dev/v2/mcp
```

A keyless Cursor or generic `mcp.json` example looks like this:

```json
{
  "mcpServers": {
    "firecrawl": {
      "url": "https://mcp.firecrawl.dev/v2/mcp"
    }
  }
}
```

If you use a key, add the Bearer header in the shape your client’s official docs show. Local runs need Node.js 22 or later:

```bash
env FIRECRAWL_API_KEY=fc-YOUR-API-KEY npx -y firecrawl-mcp
```

The README’s local example is usually unpinned. Some local docs have shown a pin such as `@3.33.2`; that is an example on that page at the time, not a version this guide is requiring. A Windows troubleshooting example uses `cmd /c "set FIRECRAWL_API_KEY=... && npx -y firecrawl-mcp"`. Do not put the key in chat or in the repository.

## A worked example

This example uses the official keyless tool names. Point it at a public page you are allowed to fetch. The tools were not run here, so this page does not invent page text.

1. Add `https://mcp.firecrawl.dev/v2/mcp` in your client, then confirm the connection with `/mcp` or the client’s MCP list.
2. Send: `Use firecrawl_search or firecrawl_scrape on https://docs.firecrawl.dev to get the current introductory description of Firecrawl. Do not invent text if the tool fails.`
3. Official docs say the keyless mode should expose `firecrawl_search`, `firecrawl_scrape`, and `firecrawl_parse`. If a tool errors or asks for a key, check whether you called crawl, map, or agent, which still need a key.
4. Open the original page and check the summary yourself. If you need a higher quota or the full tool surface, create an `fc-` key in the Firecrawl dashboard and switch to a config that sends an Authorization header.

## Common pitfalls

- Putting the API key in the URL. Official docs forbid that. Use `Authorization: Bearer`.
- Calling crawl, map, or agent in keyless mode. Those tools need a key. Keyless mode covers search, scrape, and parse only, and it is rate-limited.
- Adding a second server entry when you reconnect OAuth. Official docs say `mcp-oauth` is an MCP endpoint; do not add another item just to re-authorize.
- Running locally on an older Node.js. Official local docs require Node.js 22 or later. Cursor’s JSON shape has also varied by version; the README mentions notes around 0.45.6 and later. Use the current Cursor docs.
- Running out of plan credits. Credits are shared across products. Metering details belong on the official pricing pages.

## FAQ

### Can I use it without an API key?

Yes. `https://mcp.firecrawl.dev/v2/mcp` exposes Search, Scrape, and Parse, with rate limits. Crawl, Map, and Agent still require a key.

### Where should the API key go?

Put it in an `Authorization: Bearer <FIRECRAWL_API_KEY>` header, or in the `FIRECRAWL_API_KEY` environment variable for a local process. Do not put it in the URL, and do not send it to the model. Official keys start with `fc-`.

### Do I have to run it locally with npx?

No. Official docs recommend the hosted HTTP remote server. Local npx is useful when you want to manage the process yourself, or when you must inject the key through an environment variable. That path needs Node.js 22 or later.

### How many keyless requests can I make each day?

Official docs only say the keyless path is rate-limited. This page did not find a citable fixed daily number on 2026-11-15, so it does not invent one. If you need a stable quota, use a keyed plan and check the official site.

### Is this a community MCP or the official one?

This is the official Firecrawl repository `firecrawl/firecrawl-mcp-server`, licensed under MIT.

## Sources and license

The [firecrawl/firecrawl-mcp-server](https://github.com/firecrawl/firecrawl-mcp-server) repository is licensed under MIT. This page is a paraphrase, for CQBase — Context & Query Base, of the README, the official MCP docs, and the plan wording on [firecrawl.dev](https://www.firecrawl.dev/), checked on 2026-11-15. If endpoints, tool coverage, or prices change, the official site wins.

## Catalog record

- Author: firecrawl
- License: MIT
- Source: https://github.com/firecrawl/firecrawl-mcp-server
- Added: 2026-11-15
- GitHub stars: 7824 (captured 2026-11-15)

