Docs · MCP
MCP docs
Connect seops to Claude, ChatGPT, Gemini or Claude Code: a remote MCP server with OAuth sign-in and 45 tools. Read tools are free for every account.
Short answer. Add https://app.seops.app/mcp to your assistant, sign in with Google, ask about a site by name. There are 45 tools: 34 read, 11 write. The server ships with rules that tell the agent how to analyse.
Connect
Claude (claude.ai, Desktop)
- Settings → Connectors → Add custom connector
- URL: https://app.seops.app/mcp
- Sign in with Google when the OAuth window opens
- Ask: "Which of my sites lost clicks last week?"
Claude Code
claude mcp add --transport http seops https://app.seops.app/mcp- Run /mcp and complete the sign-in
- The server ships analysis rules and a task-sync protocol; the agent follows them without extra prompting
ChatGPT
- Settings → Connectors → Create (developer mode)
- MCP server URL: https://app.seops.app/mcp, authentication OAuth
- Enable the connector in a chat and ask about a site by name
Gemini
- Gemini CLI: add the server to settings.json under mcpServers with httpUrl
- Gemini app connectors follow when Google opens custom MCP to all accounts
Rules the agent follows
The server's MCP instructions are sent to the agent on every connection. In short:
- No guessing: every number comes with its source and period.
- Rates are 0–1 in the API; the agent is told to show percentages.
- Fewer than 50 clicks or 500 impressions in 28 days is called a small sample.
- The agent never changes panel, Cloudflare or GA4 settings; it gives copy-paste text.
- AI tasks start only after you approve them in the panel.
Tools
Read tools are free for every account. Write tools only change the panel's task list, prompt history and drafts; they cannot change settings and never cost credits.
Projects and overview
get_work_items | Data-derived to-dos with counters (baseline, remaining, done) in fixed impact order; start a session with the top open item. |
list_projects | Tracked sites, connected sources (Search Console, GA4), keyword and prompt counts. |
get_overview | 28-day summary of one project: Search Console totals, GA4 traffic and AI-assistant visits, rank movers, AI visibility rates. |
get_weekly_summary | Last week (Mon–Sun) across every module, with what changed. |
get_alerts | Open alerts: traffic drops, index losses, stale backups, failing crons. |
Search
get_search_console | Totals vs previous period, daily series, top queries and pages. |
get_search_opportunities | Queries at position 5–20 with the page that gets most impressions for each. |
get_rank_tracking | Latest position per keyword and device (Search Console), change, ranking page. |
get_keyword_history | 90-day history for one keyword. |
get_indexing | Sitemaps, URL inspection results, IndexNow key and queue. |
get_queue | Pending rank checks and AI runs, recent errors, next cron times. |
Analytics
get_analytics | GA4 totals vs previous period, channels, landing pages, source/medium, AI-assistant traffic. |
get_realtime | Active users in the last 30 and 5 minutes, by page and source. |
get_navigation | Where visitors to a page came from and went next; frequent three-step paths. |
get_site_search | Internal site search terms, zero-result searches. |
AI visibility
get_ai_visibility | Mention and citation rates per engine and prompt; cited sites. |
get_ai_answer | Full answer text of one run, cited URLs, searches the model performed. |
get_cloudflare_ai_crawlers | GPTBot, ClaudeBot, PerplexityBot and others: requests, pages, allowed vs blocked. |
Site health and Cloudflare
get_site_audit | Technical crawl of sitemap pages: status, redirects, noindex, titles, descriptions, canonicals, structured data. |
get_page_speed | PageSpeed Insights for the most important pages, mobile and desktop. |
get_cloudflare_security | Requests, visits, data, blocked requests by action, feature, rule, country and path. |
get_cloudflare_usage | Workers, D1 and R2 usage against plan limits. |
get_backups | D1, Supabase, WordPress and Hostinger backup status and last successful run. |
get_security_audit | Security check: external-surface results (HTTPS, security headers, exposed files, security.txt, SPF/DMARC, Cloudflare TLS) and the latest code audit with its findings. |
submit_security_auditwrite | Submit the findings.json of a code audit made with Cloudflare's security-audit skill; new and closed findings are matched to the previous run. |
get_waf_rules | The site's Cloudflare WAF rules with the panel's analysis: dry run against real URLs and verified bots, false positives in the last 7 days, proposals. |
save_waf_ruleswrite | Record a WAF rule proposal with its rationale, or a new public endpoint; the panel never writes to Cloudflare, the user applies it. |
get_access_audit | Who can reach the site: search and AI bots, agents, PageSpeed and visitors from other countries; WAF linter with a corrected rule set. |
get_resilience_runs | Results of controlled load tests: whether the site stayed up and how fast it answered. |
run_resilience_testwrite | Send a controlled load (at most 250 requests) to your own verified site; only when you ask and retype the domain. |
get_usage | This month's paid-provider spend, budget cap, balance, pacing. |
Tasks, prompts, marketing (write)
list_tasks | Task list filtered by project, owner and status. |
add_taskwrite | Open a task with what/why/verification; owner user or ai; dedupe key. |
update_taskwrite | Change status, priority, note; close with proof. AI tasks start only after approval in the panel. |
get_task_log | Every change on tasks: who, when, what. |
log_promptwrite | Record the user's request and the agent's summary in prompt history. |
list_prompts | Prompt history with topics and tags. |
list_prompt_templates | Prompt library. |
save_prompt_templatewrite | Save a reusable prompt to the library. |
list_social_posts | Content and sharing: accounts, posts, schedule. |
create_social_postwrite | Draft a post from the week's data; the user approves before it goes out. |
update_social_postwrite | Edit or schedule a draft. |
save_content_draftwrite | Save a guide or article draft for review in the panel. |
list_feedback | Bug reports and suggestions logged by agents. |
update_feedbackwrite | Close or annotate a feedback item. |
Auth and security
- OAuth 2.1 with PKCE; dynamic client registration is on. A token is bound to one workspace and cannot read another.
- Secrets (keys, emails) are masked server-side and never written raw to prompt history.
- Agents write each request and a short summary to Prompt history (Tracking › Tasks › Prompt history), with secrets masked; you can turn this logging off.