MCP Server

Spider runs a built-in MCP server so AI coding assistants can query crawl data directly. They read audit results, issues, Google metrics, and site structure without anything being copied out by hand.

Project binding

Every server process is locked to exactly one project. The --project argument is mandatory and the server refuses to start without it. There is no mode that reaches every project.

The binding is enforced two ways:

• Injected, not accepted — the bound project id is inserted into every project-scoped tool. Tools take no projectId parameter, so an agent has no field through which to name a different project.
• Ownership-validated — any crawlId or pageId an agent supplies is checked server-side first. A real id belonging to another project is refused.

Call get_mcp_scope at the start of a session to confirm which project the server is attached to.

Connecting a client

Spider generates a ready-to-paste configuration per client. Open a project's Connect AI agent action, or Settings → MCP Server, to copy the exact snippet.

Claude Desktop has one global config and no per-project scope, so a binding there applies to every Claude Desktop conversation.

The desktop app must be running. The access check fails closed, so if the app is starting or busy the tools report that the connection could not be verified rather than silently returning empty results.

Packaged SEO skills

The server registers 8 skills as MCP prompts. Each drives the tools below in a set sequence, so an assistant runs a whole workflow from one instruction instead of calling tools individually and stitching results together.

Tool reference

The server exposes 64 tools. They are read and safe-write only: the only tools that change anything are the crawl controls, run_workflow, and the two Knowledge proposals. No tool deletes a project or its data.

Projects and crawls

Entry points for finding a project and its crawl history.

• list_projects — every SEO project in Spider
• get_project — one project's detail and recent crawls
• list_crawls — crawl history with status and scores
• get_crawl_summary — site-wide scores, metrics, and Core Web Vitals spread
• get_crawl_status_summary — page counts grouped by each status enum
• get_crawl_logs — crawl events, warnings, and discovered URLs for debugging

Crawl control

Start and steer a crawl. Stopping keeps everything already crawled.

• start_crawl — begin an audit; devices come from the project config
• pause_crawl — pause and keep progress
• resume_crawl — continue a paused crawl
• stop_crawl — end a crawl, retaining crawled data
• wait_for_crawl — block until a crawl finishes, then return final status
• list_active_crawls — check before starting another crawl

Pages

Page-level detail, from a whole report down to a single tab.

• list_pages — crawled pages with scores, search, sorting, pagination
• get_page_report — the full report for one page
• get_page_section — one section (metadata, headings, links, schema…)
• get_page_category_detail — the per-tab payload as the app renders it
• get_page_screenshot — the URL of a page's rendered screenshot
• get_site_graph — structure, depth, orphans, indexability

Issues

Aggregated findings, plus the catalog behind them.

• list_issues — issues across the site, filterable by category and severity
• get_critical_issues — critical and high-severity findings first
• get_issue_pages — every page affected by one issue type
• get_page_issues — every issue on one page, with fixes
• get_quick_wins — high-impact, low-effort improvements by ROI
• get_broken_links — broken links with error types and source pages
• get_redirects — redirects by type, including chains
• get_duplicates — duplicate titles, descriptions, H1s, or content
• get_issue_definition — one issue's severity, impact, and recommended fix
• list_issue_definitions — the full catalog, independent of any crawl

Site-level checks

Whole-site signals rather than per-page ones.

• get_site_analysis — robots.txt, sitemap.xml, SSL/TLS, llms.txt compliance
• get_robots_blocked_urls — the actual blocked URLs and top blocked paths
• get_sitemap_coverage — orphans, and in-sitemap-but-not-crawled URLs
• get_project_health — an A+–F grade blending scores, issue load, and trend

Google data

Live OAuth connections and imported Search Console data.

• get_google_metrics — per-URL GSC or GA4 performance
• get_gsc_import_top_pages — top pages by clicks or impressions
• get_gsc_import_top_queries — top queries, site-level
• get_gsc_sitemaps — Google's own view of submitted sitemap health
• get_url_inspection — live URL Inspection: indexed state and coverage
• list_gsc_imports — Page-Indexing CSV imports and exclusion reasons
• get_ga4_sections — GA4 KPIs and section breakdowns in one call
• get_google_search_incidents — Search Status Dashboard incidents

Prioritization

Where traffic data turns a list of issues into an order of work.

• get_issues_with_traffic — issues ranked by traffic at risk
• get_high_traffic_at_risk — high-traffic pages that also have issues
• get_google_opportunities — four-layer audit, GSC, GA4 correlation
• get_link_insights — which pages should link to under-linked pages
• get_content_opportunities — new pages worth writing, with outlines
• get_backlinks — backlink profile and referring domains
• get_keywords — the keyword universe mapped onto site topics

Trends and comparison

Change over time, and across devices.

• get_project_trends — score trends across crawls
• compare_audits — two crawls side by side
• get_audit_deltas — new versus fixed issues since the last audit
• get_device_comparison — metrics across desktop, mobile, and tablet
• get_device_url_gaps — per-URL score gaps between devices

Workflows

Spider's recipes, run from the agent. Generated images need the download tool.

• list_workflows — available workflows and their input specs
• run_workflow — run one by id; returns a runId
• get_workflow_run — poll for status and results
• list_image_references — Image Studio reference images and their ids
• download_workflow_image — save a generated image locally

Knowledge base

Owner-authored business facts that ground every AI feature. Writes are review-first by design.

• get_knowledge — documents, or the exact context block AI calls receive
• propose_knowledge — propose one document; it lands as an inactive draft
• activate_knowledge — activate a draft the user confirmed verbatim in chat
• add_competitor — track and profile a competitor from their sitemaps

Connection

Diagnostics. get_mcp_status and get_mcp_scope are free and never count against usage.

• get_mcp_status — connection status and usage
• get_mcp_scope — the project this server is locked to
• get_mcp_diagnostics — detailed environment and connection diagnostics