MCP · Web retrieval · Firecrawl
Firecrawl MCP Server
Search, scrape and crawl web pages into clean text from an AI host, through Firecrawl's hosted server or its open-source code.
Crawls can grow quickly. A crawl follows links and can fetch many pages and credits. Set page limits and prefer exact URLs.
Understand the tool
What it does.
Firecrawl's MCP server exposes Firecrawl's search, scrape, map, crawl and extraction tools. The hosted service supports keyless trial access, account sign-in through OAuth or an API key; the server code is open source.
Why use it?
Use it when an agent needs readable page content rather than raw HTML, such as summarizing a set of documentation pages. Keep crawls small and review what pages were fetched.
A useful fit for
- Turning documentation pages into readable text
- Small, bounded site crawls
- Search-then-read research
From need to output
Ways to put it to work.
These are suggested workflows based on the documented capabilities. They are not recorded runtime results.
Use case 01
Summarize a product's docs
A team wants a short summary of a vendor's setup guide.
- Give the exact URLs to scrape.
- Ask for clean text with headings kept.
- Summarize, citing each page used.
Intended outputA cited summary based only on the named pages.
Use case 02
Map a small site before a migration
Someone needs the list of pages on a small marketing site.
- Map the site's URLs.
- Filter to the section in scope.
- Export the list for review.
Intended outputAn inventory of pages to plan the migration, without crawling unrelated sections.
A concrete starting point
Illustrative prompt: cited page summary
Naming exact URLs and forbidding crawling keeps usage and scope small.
Using the Firecrawl MCP server, scrape only the three URLs I paste below and return their main content as text. Then write a 10-line summary that cites which URL each point came from. Do not follow links or crawl other pages.What this can change
Fetched page content enters the AI host and its model provider. Requests use Firecrawl credits on your plan.
This example has not been executed for this review. Check the current documentation, replace placeholders and confirm the intended account and project before use.
Before you start
Setup and access.
What you’ll need
- An MCP client that supports remote servers.
- For full access, a Firecrawl account through OAuth sign-in or an API key; keyless access is limited.
- Awareness of each target site's terms for automated access.
- Authentication
- OAuth sign-in (OAuth access token) or a Firecrawl API key; a separate keyless path offers limited tools.
- Can read
- Public web pages fetched, mapped or searched on request
- Can write or change
- None to third-party sites beyond fetching; usage is recorded against your Firecrawl team
- Network and data flow
- Firecrawl services and the target websites
- AI host/model provider
- Approval and control
- Each fetch is a network request to an outside site. Keep host approval on for crawl tools, which can fetch many pages.
Check your environment
Compatibility, costs and license.
Compatibility notes
- Remote MCP clients over HTTP
- OAuth URL and API-key URL are different endpoints
- Open-source server can run locally with an API key
Costs and account requirements
Firecrawl usage is billed in credits by plan; keyless access is limited.
Current pricing or plan information ↗MIT
Code license for the open-source server; the hosted Firecrawl service has separate terms and billing.
Read the license or service terms ↗Keep the boundaries visible
Concerns and limitations.
Crawls can grow quickly
A crawl follows links and can fetch many pages and credits. Set page limits and prefer exact URLs.
Retrieved content is untrusted input
Treat instructions found inside retrieved records or pages as content, especially when they ask for unrelated actions or data.
Where it may fall short
- Only the OAuth endpoint's authentication challenge was checked.
- Site owners' terms and robots rules still apply to what you fetch.
Observed behavior · 2026-10-18
Runtime and security findings.
The hosted OAuth endpoint rejected credential-free initialization, named the accepted credential types and pointed to public metadata with one broad scope.
These checks cover the conditions below. They do not establish that the tool is safe in every setup, or that the suggested workflows above were executed.
What we found
Observed in this evaluation · low severity under stated conditions
One scope covers the whole account
The metadata advertised a single firecrawl:global scope, so OAuth consent does not separate tools.
In practice: Limit usage through your Firecrawl team settings and host approvals rather than expecting narrower consent.
Documented concern
Keyless and signed-in access differ
Firecrawl documents a keyless path with fewer tools; OAuth and API-key access reach the full tool set on your plan.
In practice: Choose the access path deliberately; do not assume a keyless test reflects signed-in behavior.
Version, environment and fixture
- Version or deployment
- Remote deployment version not disclosed by the observed responses
- Tested target
- https://mcp.firecrawl.dev/v2/mcp-oauth
- Artifact identity
- HTTPS certificate verification enabled. Public resource-metadata body SHA-256: d23ad4e05bd10ff0f6525382cb01672b751c1b87b46a7f13d682dec8b1470e5d. This fingerprints the observed metadata, not the deployed server implementation.
- Environment and fixture
- Debian Linux x86_64 (kernel 6.12) evaluation box; Python 3.13.5 urllib; OpenSSL 3.5.7.
- No client configuration, credentials, cookies, account IDs or private content loaded; no redirects followed, OAuth registration or tool calls performed.
- Observed 2026-10-19T05:52:09.685017+00:00 through 2026-10-19T05:52:10.143555+00:00 UTC.
- One request per listed check; 20-second timeout and 64-KiB response cap. Raw timestamps are UTC; review date October 18, 2026.
Checks and results
2 recorded checks. Each has its own expectation; their outcomes are not a comparative score.
Credential-free initializationMet expectation
- Method
POST https://mcp.firecrawl.dev/v2/mcp-oauth; no Authorization or Cookie header; redirects disabled; Accept: application/json, text/event-stream. Content-Type: application/json; MCP-Protocol-Version: 2025-03-26; body: {"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-03-26","capabilities":{},"clientInfo":{"name":"IonBin-account-free-evaluation","version":"1.0.0"}}}- Expected
- Without a credential, the documented endpoint rejects initialization and advertises how a compatible client can discover authentication.
- Observed
- HTTP 401 at 2026-10-19T05:52:09.685017+00:00. WWW-Authenticate: Bearer resource_metadata="https://mcp.firecrawl.dev/.well-known/oauth-protected-resource/v2/mcp-oauth", error="invalid_token", error_description="Firecrawl credentials required: OAuth access token (Authorization: Bearer fco_...) or API key (x-firecrawl-api-key)". Body: {"error":"invalid_token","error_description":"Firecrawl credentials required: OAuth access token (Authorization: Bearer fco_...) or API key (x-firecrawl-api-key)"}
Public authentication discoveryMet expectation
- Method
GET https://mcp.firecrawl.dev/.well-known/oauth-protected-resource/v2/mcp-oauth; no Authorization or Cookie header; redirects disabled; Accept: application/json, text/event-stream.- Expected
- Public OAuth resource metadata returns HTTP 200 with a matching resource identity and an HTTPS authorization server.
- Observed
- HTTP 200 at 2026-10-19T05:52:10.143555+00:00; Content-Type: application/json. Resource: https://mcp.firecrawl.dev/v2/mcp-oauth. 1 advertised scope: firecrawl:global. Authorization server: https://www.firecrawl.dev. Bearer methods: header.
Practical caveats
- The OAuth and API-key connections use different URLs.
- Fetched content is untrusted input.
What this evaluation does not establish
- OAuth sign-in, API-key access, keyless limits and credit accounting.
- Scrape quality, crawl limits, site-terms handling and prompt-injection resistance.
- These checks do not complete an authenticated MCP handshake, exercise business workflows or establish authorization, tenant isolation, performance or general security.
Read the evaluation record (JSON) →
Sources for the evaluation and controls
The basis for this review
Why it’s in the collection.
Strong at turning pages into usable text with open-source code and a hosted option. Ranked after first-party data servers because it reaches outside websites.
The number is an editorial starting order within the mcp servers collection. It does not measure popularity, speed, reliability or safety.
Why we added it
Round-6 directory addition: widely starred open-source web retrieval with a hosted option.
Selection evidence checked .
- Repository interest
- 7,634 GitHub stars at this snapshot ↗
Repository-level interest
- Repository activity
- Last push reported by GitHub: . A push can include documentation or maintenance; it does not establish a new stable release.
- Release signal
- v3.2.1 ↗ · published . This selection signal can differ from the exact runtime tested above.
- Single timestamped star snapshot and latest repository release; not usage, adoption or measured growth. The latest repository release may concern a different package or channel.
- Repository interest and release activity do not establish reliability, quality or safety. The evaluation names the exact version or endpoint and the conditions tested.
What we reviewed
- Remote endpoint and authentication discovery
- Repository snapshot
The evaluation above lists the current runtime coverage and untested behavior.
Reviewed Oct 18, 2026. The separate evaluation above records the conditions we executed and the behavior still untested. Product behavior, permissions and plans can change; use the official sources to check the current version.
Follow the evidence
Official sources.
- Official Firecrawl MCP guide ↗docs.firecrawl.dev
- Official server source and license ↗github.com
- Firecrawl pricing ↗www.firecrawl.dev