MCP access

SERPclimber exposes an OAuth-protected Model Context Protocol server at:

https://serpclimber.com/mcp

Use the MCP menu in the menubar and choose Copy MCP server URL, or open Settings → Account → MCP access (MCP → Manage connections). Paste the URL into a client that supports remote Streamable HTTP MCP servers. The client opens SERPclimber so you can sign in and approve access.

Opening the URL as a normal web page is not an MCP connection test. A healthy unauthenticated protocol request returns 401 with OAuth discovery details.

Connect a client

There is no API key to paste. The client discovers OAuth from the server URL and asks you to approve its scopes.

Claude Code

claude mcp add --transport http serpclimber https://serpclimber.com/mcp

Cursor, VS Code, and other mcp.json clients

{
  "mcpServers": {
    "serpclimber": { "url": "https://serpclimber.com/mcp" }
  }
}

For Claude Desktop, claude.ai, ChatGPT, and similar products, add a custom connector with the same URL. If asked for a transport, choose Streamable HTTP.

Scopes

• mcp reads owned projects, goals, findings, Tasks (by ID or #number) and their threads, Performance evidence, Audit results, Reports, integration status, settings, and documentation.
• mcp:write permits paid services and the guarded action workflow only on managed projects with an active plan. It does not bypass project ownership, access, scope, autonomy, spend, approval, or integration checks.

A read-only MCP connection is available on free projects: your external agent can read the backlog and saved evidence and perform work in your own stack. Posting progress or changing a Task requires mcp:write, a managed project, and the action workflow below.

Writes require two steps

MCP clients cannot send an unchecked mutation. They first call actions_prepare with the requested action. SERPclimber freezes the target, intended change, preconditions, warnings, risk, cost, recovery path, verification method, and expiry into a short-lived plan.

The client then calls actions_execute with only that plan ID. Plans are one-time, expire, and are checked again before execution. If the target or policy changed, the client must prepare a new plan. Actions requiring a person still wait for that approval; possessing mcp:write is not itself approval.

Tool discovery and project selection

Clients receive the current tool descriptions plus input and output schemas through tools/list. Use the workspace context tool first, copy the exact project ID, and pass it to project-scoped reads and writes. SERPclimber never silently substitutes another project when an explicit selection is invalid.

The operational tool surface covers the same workspace as the UI:

• the project operating picture;
• paginated Task lanes, individual Task threads, comments, assignments, and Task actions;
• Performance evidence for the SEO, GA4, AI answers, Crawlers, and Authority views (stored summaries for the 7-day, 28-day, 90-day, and all-history windows where available; live computation for filters and other windows), plus data freshness;
• Audit runs and findings;
• the latest saved crawl observation and detected issues for an exact project page URL;
• Reports and project settings;
• prepared and executed actions.

Callable workspace tools are workspace_read_context, workspace_read_billing, integrations_read_status, projects_read_overviews, project_read_operating_picture, tasks_list, tasks_read, audit_read, opportunities_summary, opportunities_list, performance_read, performance_search_summary, performance_topic_clusters, performance_traffic_breakdown, performance_ai_answers, performance_crawlers, performance_sync_status, reports_list, reports_read, authority_summary, authority_link_sources, authority_referring_domains, authority_backlinks, authority_pages, authority_anchors, authority_link_targets, docs_search, docs_read, search_console_analyze, search_console_find_page_movement, pages_list_project_inventory, pages_read_crawl, ai_visibility_read, ai_visibility_ask_engines, settings_read_project, settings_read_workspace_defaults, actions_prepare, and actions_execute.

Use pages_read_crawl with a project ID and an exact URL to inspect the last saved fetch: status, redirects, indexing directives, headings, content and link counts, and detected issues. It reports the crawl date and coverage and returns up to 100 issues, most severe first, with the full retained count. issuesComplete is false when evidence is partial, still being collected, unavailable, or truncated; zero recorded issues is not a clean bill for the page. “Not crawled” means there is no saved record for that exact URL; it does not prove the page is absent. Expired detail is unavailable even if the inventory still remembers the page, and checks absent from older observations are reported as unknown. This read never fetches the page or starts a crawl.

Results remove credentials, encrypted values, OAuth tokens, and provider secrets. Internal support, research, content-writing, client-only, and marketing administration capabilities are not advertised over MCP.

Manage or revoke a client

Open Settings → Account → MCP access to see each client, its scopes, connection date, last use, and call totals. Choose Revoke to invalidate its access and refresh tokens immediately.

Troubleshooting

• If sign-in never opens, confirm the client supports remote Streamable HTTP, OAuth discovery, and dynamic client registration or client metadata documents.
• A 401 usually means the client should reconnect and obtain a new token.
• A 403 insufficient_scope means the requested tool needs mcp:write.
• Project, subscription, usage, approval, or integration errors should be fixed in SERPclimber; recreating OAuth will not change them.
• The protected-resource metadata at https://serpclimber.com/.well-known/oauth-protected-resource/mcp should name the MCP resource and authorization server.