SEO

Screaming Frog MCP: set up the SEO Spider's server

Screaming Frog SEO Spider v24 added a built-in MCP server for crawling and analyzing sites from Claude, Cursor, LM Studio and other MCP clients. Here is how to set it up, fix the common errors, and run the community headless server for unattended audits.

Jesse Sibley
By , FounderPublished · Last updated · 12 min read

This guide covers which Screaming Frog version you need, how to enable the server and connect it in Claude Desktop, Claude Code, Cursor and LM Studio, why ChatGPT is harder, the community headless option, and example prompts for technical audits. If you cannot run the app locally, there is a hosted option at the end.

What is Screaming Frog MCP?

Screaming Frog MCP is the bridge between your AI agent and the Screaming Frog SEO Spider app. You ask a question in plain language, and the agent picks the Screaming Frog tools to call. Screaming Frog announced in their v24 release post (May 19, 2026) that the MCP server lets you "run crawls, analyse, export and manipulate data using the SEO Spider and node.js within Claude, LM Studio and other AI chat assistants."

Screaming Frog's MCP Server guide documents 29 tools. They cover crawl control (start, pause, resume, progress tracking), reports and bulk exports with field selection, URL-level inspection, screenshots and embeddings exports. An optional Node.js script runner, with npm and filesystem read and write tools, handles post-processing.

Who Screaming Frog MCP is for

Screaming Frog MCP fits when:

  • You already pay for Screaming Frog and want to use that data in conversation with your agent
  • You want full control over crawl configuration, JavaScript rendering and custom extraction
  • You are comfortable leaving the desktop app running while you work
  • You need interactive session features like visualizations, screenshots and crawl comparison

It may not fit when:

  • You want to run scheduled audits with the GUI closed (see the community headless option below)
  • You want to use it from ChatGPT or claude.ai on the web

Step 1: check your version and licence

Download the latest SEO Spider from screamingfrog.co.uk/seo-spider. The v24 release added the MCP server, and Screaming Frog's FAQ says to install v24.1 or later before the Claude Desktop extension: installing the extension first, or on v24.0, can fail with an "Unexpected non-whitespace character" error.

The MCP server is a paid licence feature: Screaming Frog's docs say it will not work in the free version. A paid licence is $279 per licence per year (checked October 2026). It also needs database storage mode (File → Settings → Storage Mode), which is the default.

Step 2: enable the MCP server

In Screaming Frog:

  1. Open File → Settings → MCP Server
  2. Accept and enable the Node.js runtime environment
  3. Click Start MCP Server. A "MCP Server Active" message appears in the bottom left of the app, and the server listens at http://localhost:11435/mcp

That is the Streamable HTTP mode, and it only runs while the SEO Spider is open: close the app and the server stops. There is also a STDIO mode, where Claude Desktop launches the Spider headless itself (see Step 3).

Step 3: connect in Claude Desktop

Screaming Frog ships two Claude Desktop extensions on its MCP Server guide: spider-mcp.mcpb for STDIO, where Claude launches the Spider headless, and spider-streamable-mcp.mcpb, which connects to the server you started in Step 2 and needs Node.js installed.

  1. Download the extension you want from that guide
  2. In Claude Desktop, open Settings → Extensions, click Advanced settings, then Install Extension
  3. Select the downloaded .mcpb file and click Install
  4. Quit Claude fully (from the tray icon on Windows) and reopen it. The extension shows as running under Settings → Developer

The extension assumes the SEO Spider is in its default install location. If it is not, edit the .mcpb file to point at your path, or reinstall the Spider in the default location.

Step 4: connect in Claude Code

Add the server from your terminal:

Terminal
claude mcp add --transport http screaming-frog http://localhost:11435/mcp

Then run /mcp in Claude Code and enable the Screaming Frog server.

Step 5: connect in Cursor

In Settings → Cursor Settings → Tools & MCP, add the server URL:

JSON
{
  "mcpServers": {
    "screaming-frog": {
      "url": "http://localhost:11435/mcp"
    }
  }
}

Any client on the same machine that supports Streamable HTTP servers can point at the same URL.

Screaming Frog MCP in LM Studio

Screaming Frog's guide also covers LM Studio, which it calls a good free alternative with local models. In LM Studio, open the Developer tab, then Local Server, and click mcp.json. Add an entry for the Screaming Frog server with the URL http://localhost:11435/mcp, save, and set the server to running. Restart LM Studio, then enable the MCP server under the tool icon in the chat.

Can you use Screaming Frog MCP with ChatGPT?

Not directly. Screaming Frog does not document a ChatGPT setup. ChatGPT adds custom MCP servers in developer mode (Settings → Apps → Advanced settings), but it connects to them from OpenAI's side over a public HTTPS URL, so it cannot reach http://localhost:11435/mcp on your machine. The same goes for claude.ai on the web.

Getting there means putting a tunnel in front of the local server, and Screaming Frog's licence terms make you responsible for making sure nobody but the licensed user can reach it. For most people, Claude Desktop, Claude Code or Cursor on the same machine is the simpler route.

Common problems and fixes

The desktop app must be running

Symptom: connection fails or tools report errors.

Cause: in Streamable HTTP mode, the built-in MCP server only runs while the SEO Spider is open and the server is started.

Fix: open the SEO Spider, click Start MCP Server in File → Settings → MCP Server, and keep the app open while you work with your agent.

"Unexpected non-whitespace character" after installing the Claude extension

Cause: the extension was installed before v24.1, or v24.0 is installed, which Screaming Frog says has a bug with some non-English languages.

Fix: uninstall the extension, quit Claude fully, install SEO Spider v24.1 or later, then install the extension again.

No paid licence

Symptom: the MCP server is not available.

Cause: the MCP server is a paid licence feature and does not work in the free version.

Fix: buy a licence from screamingfrog.co.uk/seo-spider/pricing.

The community headless option: bzsasson/screaming-frog-mcp

The bzsasson/screaming-frog-mcp server is a community-built Python wrapper around the Screaming Frog CLI. It works with the GUI closed for scheduled audits, CI pipelines and unattended agents.

How it differs from the official server

The official server has 29 tools, runs as the SEO Spider itself for the whole session, and offers visualizations, screenshots and optional Node.js scripting. The bzsasson server has 9 read-and-export tools, works with the GUI closed, and is built for runs nobody is watching.

FeatureOfficial built-in MCPbzsasson headless server
Tool count299
GUI requirementApp open in HTTP mode; STDIO mode launches it headlessGUI must be closed (database lock)
SetupStart in File → Settings → MCP Server, or STDIO mode via Claude Desktop extensionPython package via pip or uv (Python 3.10+)
Node.js scriptingYes (optional, with filesystem access)No
ScreenshotsYesNo
Best forInteractive sessions with visualizationsScheduled audits, CI, unattended agents

Install the headless server

  1. Install it as a persistent uv tool (or with pip install screaming-frog-mcp):
Terminal
uv tool install screaming-frog-mcp
  1. Find the executable with which screaming-frog-mcp, usually ~/.local/bin/screaming-frog-mcp.
  2. Add it to your MCP client with the full path. On Linux or Windows, also set SF_CLI_PATH to the Screaming Frog CLI launcher:
JSON
{
  "mcpServers": {
    "screaming-frog": {
      "command": "/path/to/screaming-frog-mcp",
      "args": [],
      "env": {
        "SF_CLI_PATH": "/path/to/ScreamingFrogSEOSpiderLauncher"
      }
    }
  }
}

The README warns against launching it with uvx in a client config: on a cold cache, the download can run past the client's start-up timeout. See the repository README for source installs and every setting.

Both need a licensed Screaming Frog install on the same machine: the official server is paid only, and the community server's README says most features need a paid licence. So crawl interactively in the GUI with your full config, close it, and let the headless server handle the unattended side.

Example prompts

Here are prompts you can try once Screaming Frog MCP is connected.

"Crawl example.com and list all broken links with their source pages and anchor text."

Redirect chains

"Find all redirect chains longer than 2 hops and show the full path for each."

Missing meta descriptions

"List all pages with no meta description, grouped by URL folder."

Duplicate titles

"Find pages with duplicate title tags and show the URLs and title text."

Non-indexable pages

"Show me all pages blocked by robots.txt, noindex meta tags or X-Robots-Tag headers."

Deep pages

"List indexable pages with a crawl depth over 4, with their unique inlinks, sorted by depth."

Orphan pages

"Identify orphan pages (pages with no internal links pointing to them) and export the list with their URLs and titles."

Screaming Frog only finds orphan pages from URLs it learns outside the crawl, so include your XML sitemap or connect Google Analytics or Search Console in the crawl, and run crawl analysis before you ask.

Run this in your agent

Connect Manifold to Claude, ChatGPT or Cursor, then paste this prompt.

Crawl example.com and list broken links, missing meta descriptions, pages with no h1, and pages returning 404, with source pages where relevant.

Limits of the local approach

The built-in Screaming Frog MCP and the community headless server both run locally on your machine. This means:

  • The desktop app and your machine must be on. If you close your laptop or quit the app, the server stops. For the bzsasson server, the GUI must be closed, but the machine must stay on.
  • Crawl size limited by your machine and licence. The free version caps crawls at 500 URLs. Paid licences allow unlimited crawls, but the maximum depends on your allocated memory and storage.
  • Harder to share across a team. Each team member needs their own Screaming Frog licence and local setup. Crawl results live on individual machines unless manually exported and shared.

Hosted option: Manifold

Manifold is a hosted SEO MCP server, so it works from ChatGPT and claude.ai on the web with no local app or Screaming Frog licence. For technical SEO it has a site crawl and a page snapshot:

  • seo_run_technical_crawl: a full-site crawl with optional JavaScript rendering. It returns an on-page score and flags broken links, duplicate titles, missing meta descriptions, non-indexable pages, redirects, render-blocking resources and low content rate, with up to 10 sample URLs per issue.
  • seo_get_page: a snapshot of one URL, with its title, meta description, canonical, robots directives, headings, word count, schema types, link counts and images without alt text.

On October 5, 2026, we ran seo_run_technical_crawl on www.manifoldmcp.com with a 200-page cap and no JavaScript rendering. It crawled 71 pages in about 4 to 5 minutes and flagged render-blocking resources on 71 pages and low content rate on 55, with 0 broken links and 0 duplicate titles.

Screaming Frog has far deeper configuration, custom extraction and exports. Manifold's crawl returns at most 10 sample URLs per issue, has no custom extraction or schema validation, and does not run Lighthouse. If you need those, Screaming Frog is the better fit.

Manifold fits when you want a crawl from any client without a local install, paid per call, next to AI answers, keyword data and Search Console data from its Search Console MCP server.

Which to choose

Comparison of Screaming Frog built-in MCP, community headless server, and Manifold hosted alternative
Screaming Frog MCP vs community headless vs Manifold: where it runs, what you need, cost model, and best fit for each option.
Screaming Frog built-in MCPbzsasson headless serverManifold
Where it runsLocal server at http://localhost:11435/mcpLocal Python packageHosted by Manifold
What you needv24.1+ with the app open (HTTP) or the Claude Desktop extension (STDIO)Python 3.10+ and the GUI closedNo local app or licence
LicencePaid licence requiredFree version capped at 500 URLs, paid needed for most featuresNo Screaming Frog licence
Cost$279/year for paid licence$279/year for paid licencePay per call; 250 free credits a month and Pro from $29/mo
Tool count299A site crawl and a page snapshot
ChatGPT/claude.ai webNo (local only)No (local only)Yes
Best forInteractive sessions with full Screaming Frog config and visualizationsScheduled audits, CI, unattended agents with GUI closedCrawls from any client, next to Search Console, AI answers and keyword data

Pick by what you already pay for and how you want to manage local apps.

You already have Screaming Frog and want to use it in conversation. Use the built-in MCP. Start it in File → Settings → MCP Server, connect via http://localhost:11435/mcp or the Claude Desktop extension, and keep the app open while you work.

You want scheduled audits or CI with Screaming Frog and the GUI closed. Use the bzsasson headless server. Install it with uv or pip, crawl in the GUI with your full config, close the GUI, and let the headless server handle exports and analysis.

You want hosted crawl tools with no local app or licence needed. Use Manifold, and pair the crawl with Search Console data in the same chat.

Questions people ask

Does Screaming Frog MCP need a paid licence?

Yes. Screaming Frog's docs say the MCP server is a paid licence feature and will not work in the free version. A paid licence was listed at $279 per year in October 2026.

How do I install the Screaming Frog MCP extension in Claude Desktop?

Install SEO Spider v24.1 or later first. Download spider-mcp.mcpb (STDIO) or spider-streamable-mcp.mcpb from Screaming Frog's MCP Server guide, then in Claude Desktop open Settings, Extensions, Advanced settings, Install Extension, pick the file, and restart Claude. The extension assumes the Spider is in its default install location.

What port does Screaming Frog MCP use?

In Streamable HTTP mode the server listens at http://localhost:11435/mcp while the SEO Spider is open and the server is started. Point Claude Code, Cursor or LM Studio at that URL.

Can I use Screaming Frog MCP from ChatGPT or Claude on the web?

The built-in server runs locally at http://localhost:11435/mcp, so it only works from clients on the same machine. For a server that works from ChatGPT or Claude on the web, use a hosted option such as Manifold.

What is the difference between the official and community Screaming Frog MCP servers?

The official built-in server has 29 tools, runs as the SEO Spider itself, and offers visualizations, screenshots and optional Node.js scripting. The community bzsasson server is headless with nine read-and-export tools, works with the GUI closed, and is built for scheduled audits and CI with a minimal tool surface.

Is there a hosted option for crawling from Claude?

Manifold is a hosted SEO MCP server with seo_run_technical_crawl and seo_get_page tools. It runs crawls without a local app or Screaming Frog licence and returns up to 10 sample URLs per issue, but has no custom extraction and does not run Lighthouse.

Sources

Jesse Sibley

Jesse Sibley

Founder of Manifold. A growth engineer who has worked with London startups for eight years.

All SEO guides
  • 16 min read

    Find keyword cannibalization with Claude

    Connect Search Console, group query and page rows by query, check the live SERP and both pages, pick one owner per query, then track with a weekly rank monitor that reports when a different page ranks.

  • 6 min read

    Claude plugins vs skills vs MCP

    What a Claude plugin, skill, connector and MCP each add, how those words work in ChatGPT, and one marketing job with Manifold MCP.