Skip to content
All documentation
Documentation · MCP
Summarise with

Give your AI client your visibility data

CiteAgentic runs a Model Context Protocol server, so Claude, Cursor and any other MCP client can read your scans, prompts, competitors and recommendations — and run new scans — without you copying anything across.

What is the CiteAgentic MCP endpoint?

It is POST https://citeagentic.com/mcp. The server speaks JSON-RPC over Streamable HTTP at that single endpoint:

POST https://citeagentic.com/mcp
Authorization: Bearer <your CiteAgentic API key>

GET /mcp answers 405 on purpose — this is a POST-only transport.

One key works for both the MCP server and the REST API — there is one credential, not two. See Authentication for how to get one and how scopes work.

What tools does the MCP server expose?

Twenty-seven tools, grouped by what they touch.

AreaTools
Sitescite_list_sites, cite_get_portfolio_summary, cite_get_site_brief
Scanscite_get_latest_scan, cite_get_scan_status, cite_run_visibility_scan, cite_run_audit_scan
Auditcite_get_audit_scores, cite_get_audit_findings
Visibilitycite_get_visibility_scores
Promptscite_list_prompts, cite_get_prompt_detail, cite_generate_prompts, cite_refine_prompt, cite_get_prompt_competitor_results
Competitorscite_list_competitors, cite_add_competitor, cite_get_competitor_gaps, cite_run_competitor_comparison
Recommendationscite_get_recommendations
Contentcite_get_content_opportunities, cite_generate_content_draft, cite_get_voice_profile
Buyer evidencecite_get_review_themes, cite_get_buyer_criteria, cite_get_positioning_claims
Agentcite_ask

What scopes does each tool need?

Each tool declares the scope it needs, and a key only carries the scopes you grant it. Read and write are separated, and so is running work that costs credits.

ScopeGrants
read:sitesSite list, portfolio summary, site brief
read:visibilityVisibility scores and history
read:auditsAudit scores and findings
read:promptsPrompt list and per-prompt detail
read:competitorsCompetitor list and gap analysis
read:recommendationsThe recommendation queue, including fix prompts
read:contentContent opportunities and drafts
read:voiceVoice profile, review themes, buyer criteria
write:promptsGenerate and refine prompts
write:competitorsAdd a competitor
write:contentGenerate a content draft
run:visibilityStart a visibility scan
run:auditsStart an audit scan

Grant the narrowest set that does the job. A key that only needs to report on visibility does not need run: scopes, and a scan started by mistake spends credits.

What should I know before wiring up MCP?

Scans are rate-limited per site. A visibility scan started through MCP respects a one-hour cooldown, the same as one started in the app. A second request inside that window returns the existing run rather than starting a duplicate and billing you twice.

Write tools accept an idempotency key. Pass idempotencyKey on a tool that creates something and a retry returns the original result instead of creating a second copy. Worth using: agent loops retry more than people do.

Every call is recorded against the key that made it, so you can see which client did what.

Quotas are the plan's quotas. Prompts, competitors and drafts count against the same limits as the UI. MCP is another way into the same account, not a way around it.

Frequently asked questions

Which AI clients work with this?
Any MCP client that supports Streamable HTTP with a bearer token — Claude Code, the Claude desktop app, Cursor, and others. The server advertises protocol version 2025-06-18.
Is the MCP key the same as the API key?
Yes. One key in Account → API keys works for both POST /mcp and the REST API. Scopes govern what it can reach.
Can an MCP client see other people's data?
No. A key resolves to its owner, and every tool is scoped to the sites that user owns or collaborates on. Row-level security enforces this in the database, not only in the route.
What happens if my plan does not include API access?
Every call returns tier_disallows_api. Nothing partially works.