Screaming Frog MCP Server.

What the Screaming Frog MCP Server is

SEO Spider 24.0, codenamed “bolus” and released on 19 May 2026, shipped the feature many of us had been waiting years for: an official MCP server. MCP is short for Model Context Protocol, an open standard that lets an AI assistant such as Claude call outside tools. Here, the tool is SEO Spider itself.

The practical upshot: you stop pasting exports and CSV files into a chat window. The assistant starts the crawl, opens saved crawls, produces reports and bulk exports, merges data sets and processes them with Node.js scripts, and you drive the whole thing in plain language. As Chris Long put it in his launch-day post, the real gain is owning the process end to end, from crawl to analysis, with no manual hand-offs along the way.

Before you start

• SEO Spider 24.0 or later. Check your version under 1Help›2About (on a Mac, 1Screaming Frog SEO Spider›2About Screaming Frog SEO Spider) and update if needed: your existing licence keeps working.
• An active licence: MCP is a paid feature and is not available in the free version.
• Database Storage mode, which is the default anyway: you can confirm it in 1File›2Settings›3Storage Mode.
• An MCP client: Screaming Frog documents Claude Desktop and LM Studio, but any MCP-capable client will do, Claude Code included.
• Node.js, only if you go for Streamable HTTP mode (more on that below).

What the AI can do with SEO Spider

Almost everything you can reach in the interface is available to the model: reports, bulk exports, the filters of every tab, single-URL data, stored screenshots and page content. On top of that, the server bundles two families of tools that make it self-sufficient:

• Node.js tools: the model can write and run scripts and pull in whatever npm packages it needs to crunch data, draw charts or build HTML pages.
• File tools: it reads, writes and organises files inside a dedicated folder, so heavy reports land on disk rather than in the conversation.

In other words, you don’t need to bolt on separate MCP servers for scripting or file access: it’s all in the box.

Screaming Frog is clear on one point, and I agree: an AI doesn’t replace an experienced SEO. It takes the grind out of repetitive work, but reading the data and deciding what matters is still your job.

Setting up the MCP server in SEO Spider

The MCP Server settings

Everything lives under 1File›2Settings›3MCP Server (on a Mac, 1Screaming Frog SEO Spider›2Settings›3MCP Server).

123456

1MCP ServerThe settings entry for the MCP server.2Start MCP ServerLaunches the server in Streamable HTTP mode with the interface open.3PortWhere the server listens: 11435 by default, the value your clients need.4DirectoryThe folder the AI reads from and writes scripts and exports to.5Node.js Runtime EnvironmentYour permission for the AI to run Node.js scripts.6View ToolsLists the tools exposed to the AI.

Here’s what each option does:

• Start MCP Server: fires up the server while the interface is open (Streamable HTTP mode). Once it’s running, “MCP Server Active” shows in the bottom-left corner. From 24.1 you can also have it start automatically every time the app opens, which is handy when SEO Spider runs on a cloud machine.
• Port: the port the server answers on, 11435 out of the box.
• Max Response Size (Bytes): the biggest single answer sent back to the model, 100,000 bytes by default. It caps how much data lands in the chat in one go.
• Directory: where the model keeps scripts, temporary files, exports and results. It can’t touch anything outside this folder.
• I understand the risks, enable Node.js Runtime Environment: lets the AI execute JavaScript and install libraries with npm. It’s off by default and you need it on in both modes.
• View Tools: opens the full list of tools the server offers.

Don’t skim past the yellow warning: turning Node.js on means allowing arbitrary code to run on your machine. Only do it for an AI client you trust.

STDIO or Streamable HTTP: which mode to pick

The server runs in one of two modes, and the choice shapes how you work:

	STDIO	Streamable HTTP
Who launches SEO Spider	The AI client, in the background (headless)	You, with the interface open
Interface visible	No	Yes: you watch crawls and data the AI opens, live
Server start	Automatic	Start MCP Server, or the MCP menu at the top
What you need	Node.js enabled in settings + STDIO extension	Node.js enabled + Node.js installed + Streamable extension
Best for	Automation and “set and forget” use	Keeping an eye on every step the AI takes

In STDIO mode the model won’t load anything on its own: ask it to list recent crawls, or give it the ID of a saved crawl, which you’ll find by right-clicking the crawl in 1File›2Crawls.

If you like watching the data in the interface while Claude works on it, Screaming Frog suggests Streamable HTTP. It’s my default too: seeing what the AI is actually doing saves a lot of surprises.

Connecting Claude

The quickest way in is Claude Desktop, Anthropic’s app for Windows and macOS, using the official extension (a .mcpb file) Screaming Frog provides for each mode. Claude does a great job but gets through tokens fast, so free plans hit their limits early.

STDIO mode

1. In SEO Spider open 1File›2Settings›3MCP Server, tick the Node.js Runtime Environment consent and confirm. No need to start the server: Claude will do that.
2. Download the STDIO extension (spider-mcp.mcpb) from the MCP Server section of Screaming Frog’s user guide, using the Download STDIO MCP Extension button.
3. Open Claude Desktop, go to 1Settings›2Extensions and click Advanced settings.

123

1ExtensionsThe desktop app section for extensions.2Advanced settingsOpens the options to install an extension from a file.3Browse extensionsThe public extension directory: the Screaming Frog MCP is installed from a file instead.

4. Click Install Extension and pick the spider-mcp.mcpb file you downloaded. Leave Use Built-in Node.js for MCP switched on, so Claude falls back on its own Node.js if the system one is missing or out of date.

123

1Use Built-in Node.js for MCPClaude uses its own Node.js when the system version is missing or outdated.2Detected toolsThe Node.js versions found on your machine.3Install ExtensionSelect the .mcpb file downloaded from Screaming Frog.

5. In the confirmation dialog, check that the author is Screaming Frog and that all requirements are met, then click Install.

1234

1InstallConfirms the installation.2Security warningThe extension can reach your whole computer: only download it from Screaming Frog’s site.3All requirements metClaude has checked prerequisites such as Node.js.4VersionThe version of the installed extension.

6. Quit Claude completely and reopen it. On Windows, closing the window isn’t enough: find Claude’s icon in the notification area next to the clock, right-click it and pick Quit, or kill the process from Task Manager. Otherwise Claude and the MCP server keep running in the background.

The extension assumes a standard install of SEO Spider, in C:\Program Files (x86)\ on Windows or /Applications/ on macOS. If you installed it somewhere else, edit the path inside the .mcpb file.

Streamable HTTP mode

1. In 1File›2Settings›3MCP Server click Start MCP Server and tick the Node.js Runtime Environment consent. “MCP Server Active” appears in the bottom-left corner. In this mode you can also start and stop the server from the MCP menu at the top.
2. Install Node.js on your computer: the Streamable extension relies on a small Node.js bridge (a shim).
3. Get the Streamable extension (spider-streamable-mcp.mcpb) from the Download Streamable MCP Extension button in the user guide and install it in Claude Desktop exactly as above: 1Settings›2Extensions›3Advanced settings›4Install Extension.
4. Quit Claude completely and reopen it.

From now on, every action the AI takes, such as opening a crawl, also shows up in the SEO Spider interface.

Checking the extension is running

Once Claude is back up, go to 1Settings›2Developer: under Local MCP servers you should see the Screaming Frog SEO Spider MCP Extension tagged running.

123456

1DeveloperThe section listing local MCP servers.2Edit ConfigOpens the MCP servers configuration file.3runningThe server is up and Claude can use it.4CommandThe SEO Spider CLI executable Claude launches.5--mcp-stdio-serverThe argument that starts SEO Spider as an MCP server in STDIO mode.6View LogsServer logs, your first stop when something breaks.

If it isn’t running, open View Logs first: the usual suspects are a non-default SEO Spider install path, an inactive licence or Node.js not enabled in the settings.

Claude Code

If you live in the terminal with Claude Code, Anthropic’s developer assistant, you don’t need an extension at all: it talks to the server directly over Streamable HTTP.

1. In SEO Spider open 1File›2Settings›3MCP Server, click Start MCP Server and enable Node.js.
2. Register the server from your terminal with a single command:

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

3. Start Claude Code and type /mcp: screaming-frog should be listed as connected.

Screaming Frog doesn’t document this setup, but it uses the same address as the Streamable HTTP server. If you changed the port, change it in the command too. The bonus with Claude Code is that generated scripts and reports land straight in your project folder, ready to version or reuse.

Connecting LM Studio and local models

LM Studio with a model running on your own machine is the free alternative to Claude. Crawl data never leaves your computer, but the quality of the analysis depends heavily on the model and the hardware.

Editing mcp.json

LM Studio connects in Streamable HTTP mode, so click Start MCP Server in SEO Spider first. Then:

1. In LM Studio click the Developer icon in the left sidebar and open the Local Server tab.
2. Hit the mcp.json button at the top centre and add the Screaming Frog server under mcpServers:

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

3. Save, close the dialog and flip the status from Status: Stopped to Status: Running.
4. Restart LM Studio, open a chat and choose a model. Click the Tool icon at the bottom right and switch on mcp/screaming-frog-mcp-server.

If you changed the port in SEO Spider, update the URL as well.

Hardware for local models

• A MacBook Pro with 64 GB of unified memory copes with models of up to roughly 40 GB.
• With a smaller GPU or less memory, expect some trial and error before you find a model that works.
• The model must support tool calling, or it simply can’t use the MCP server.
• Bigger models and reasoning (thinking) models give better answers, especially to complex prompts.
• Overall, Claude is still more dependable than a local model.

First prompts to try

Crawl and summarise the issues

Start simple: ask Claude to crawl a site and rank the main problems by priority.

1234

1The promptCrawl plus issue summary in a single request.2Used seospider integrationClaude is calling the Screaming Frog MCP tools.3High PriorityIssues grouped by priority, as in the Issues tab.4ModelThe Claude model used for the conversation.

Use Screaming Frog to crawl www.example.com, then give me the most important issues it found, ordered from highest to lowest priority.

Keep tab, filter and report names exactly as they appear in the interface: the model maps them straight onto the right tools.

Exports, scripts and an HTML report

Things get interesting when the job takes several steps. In Screaming Frog’s demo, Claude opens the latest crawl of a site, finds the bulk export for images missing alt text, runs it and builds an HTML page showing every image in a grid.

12345

1The promptOpen the latest crawl, export images without alt text and build an HTML page.2Crawl IDClaude identifies the latest crawl of the site by its ID.3Sf load crawlThe tool that loads the saved crawl.4Images Missing Alt Text InlinksThe bulk export the AI picked from those available.5Sf generate bulk exportThe tool that writes the export to a file.

The output is an interactive HTML page, saved in the MCP folder and opened in the browser, with totals, tabs, a filter and the inlink count for every image.

1234

1SummaryUnique images, total inlinks and a split by type.2TabsContent images kept apart from avatars and gravatars.3Filter and sortSearch by URL or inlink and sort by inlink count.4InlinksHow many pages link to each image missing alt text.

Going further: internal link equity

The showpiece of the launch was an internal link equity analysis. Starting from a crawl that already had Ahrefs metrics pulled in as columns, Claude builds the internal link graph, calculates a weighted internal PageRank, rolls the scores up by site section and draws an interactive chart of how equity flows between them.

123

1Key figuresURLs in the graph, unique edges and how much equity stays in or leaves each section.2LegendRed marks the sections where outside authority enters (Ahrefs).3/blog/One of the sections taking in the most outside authority.

Load my saved Screaming Frog crawl called 'example', which already includes Ahrefs URL Rating, Backlinks and Referring Domains columns. Build the internal link graph, run a weighted internal PageRank, group the results by site section and produce an interactive chart showing how link equity moves between sections.

More prompts to adapt

A few more ideas, written my way, to tweak for your own site:

Crawl www.example.com, work through the Issues Overview report, rank the problems against current technical SEO best practice and turn everything into a detailed HTML report backed by the relevant bulk exports.

Export the broken internal links (client errors) and cross them with the Internal tab, using GA Sessions, Search Console Clicks and Unique Inlinks to decide which fixes come first.

Load the two most recent crawls of www.example.com, compare their main issues and tell me what changed that actually matters.

Export every image heavier than 100 KB and build an HTML gallery from that file, with each image's URL underneath, a title and a short intro explaining the page. Open it in my browser.

Create the Redirect Chains report and visualise the chains, flagging any whose final destination is non-indexable.

Put every non-200 URL, together with its inlinks, into a single CSV file.

From the Internal tab, HTML filter, export only Address and Unique Inlinks to all_urls.json, find the 20 URLs with the most inlinks and chart them as a bar graph in an HTML page.

Export Address and CO2 (mg) for all HTML URLs to co2.csv, then use a script to list the 10 pages with the highest CO2 values.

List the images that are missing width and height attributes.

Early community tests have added a few more angles: spotting duplicate H1s across a site, finding orphan pages with internal linking opportunities, and checking crawl depth against organic landing pages from Search Console. Pair the MCP with the API integrations (Search Console, Analytics, Ahrefs) and you get complete workflows for audits, internal linking and content monitoring.

Further readingOrphan pages.How to find orphan pages by combining the crawl, XML sitemap, Analytics and Search Console: a method you can now hand to an AI too.Read the guide →

The tools the MCP exposes

This is the vocabulary the AI uses to talk to SEO Spider. Some clients, LM Studio among them, show tool calls as they happen, and reasoning models even explain why they picked a given tool, which helps when something goes wrong. The up-to-date list is always one click away under View Tools in the settings.

Crawls and saved crawls

Tool	What it does
sf_list_crawls	Lists recent crawls, newest first (10 by default)
sf_load_crawl	Loads a saved crawl by its ID
sf_crawl	Starts a crawl from a URL, with optional config, project and name
sf_crawl_progress	Reports progress on the running crawl
sf_pause_crawl / sf_resume_crawl	Pauses and resumes the crawl
sf_clear_crawl	Clears a paused crawl
sf_export_crawl	Exports the loaded crawl to a file

Reports, exports and URL data

Tool	What it does
sf_list_available_reports / sf_generate_report	Lists and generates reports, as CSV or NDJSON
sf_list_available_bulk_exports / sf_generate_bulk_export	Lists and generates bulk exports
sf_export_seo_element_urls	Exports the URLs of a tab and filter, with the fields you choose
sf_list_available_filters_for_seo_element	Lists the filters available for each tab
sf_list_available_data_fields_for_seo_element_and_filter	Lists the exportable columns for a tab and filter
sf_bulk_export_page_content	Exports visible text or raw HTML for every page
sf_export_embeddings	Produces a CSV of URLs and embeddings
sf_url_info / sf_url_content / sf_url_links	Data, content and inbound or outbound links for one URL
sf_get_url_screenshot	Retrieves a page’s stored screenshot
sf_open_url_in_browser	Opens a URL or a file in the browser

Node.js and file system

Tool	What it does
sf_run_node_js_script	Runs a Node.js script
sf_npm_install	Installs an npm package
sf_read_text_file / sf_write_text_file	Reads and writes text files (writing overwrites without asking)
sf_list_allowed_base_directory	Shows the folder the server is allowed to use
sf_list_directories / sf_create_directory	Lists and creates folders

The Node.js tools stay off until you give consent in the settings. If you work with embeddings, sf_export_embeddings saves you a lot of effort.

Further readingSEO Embeddings with Screaming Frog.What embeddings are, how to generate them in SEO Spider and how to use them for semantic similarity, clustering and internal linking.Read the guide →

Security, privacy and limits

Licence terms and data privacy

By running an MCP server you commit to making sure nobody other than the licensed user uses the software: don’t expose the server to other people or to your network. Whatever data flows through the MCP, in either direction, its privacy is on you. With Claude, crawl data goes through Anthropic’s servers, so think twice before analysing client sites with confidential data. With LM Studio and a local model it stays on your machine.

Scripts and Node.js

Letting a language model run arbitrary scripts is hugely powerful and genuinely risky: a malicious script could read local data, reach the network or keep running after the session ends. Screaming Frog limits file access to the MCP folder, but the advice stands: enable Node.js only for an AI client you trust, and only install extensions from the official site.

Context window: one job per chat

Large reports and big crawls blow through a model’s context window quickly. That’s exactly why the MCP can write reports to files and process them with scripts instead of dumping everything into the chat. Two habits that help:

• start a fresh chat for each job, so the conversation doesn’t fill up with stale data;
• ask the AI to save exports to files and work on them with a script, especially past a few thousand URLs.

What it can’t do (yet)

Not every SEO Spider feature is reachable through the MCP yet: Screaming Frog plans to add more based on user feedback. And the AI does make mistakes, so always check the numbers in the interface before they go into a report or in front of a client.

Further readingScheduled crawls.Need recurring crawls without AI? Scheduling, automatic exports and, since version 24, automatic comparison of the last two crawls.Read the guide →

Video tutorial

A five-minute walkthrough of the Claude Desktop setup by Usman Akram. You can also watch it on YouTube.

Video tutorialHow to Set Up Screaming Frog MCP Server Inside Claude (in 5 minutes)Watch the video5:04