CitationBench's hosted MCP server gives you ~35 SEO and GEO tools via one endpoint. Build an SEO agent with MCP for any client without months of integration work.
# Add CitationBench to your MCP client> claude mcp add citationbench --url https://mcp.citationbench.com/mcp✓ Connected · 35 tools available# Run keyword research from your agentresearch.keyword.research> { "seed": "seo automation tools", "limit": 20, "label": true }{ "keywords": [{ "kw": "seo agent api", "intent": "COMMERCIAL", "relevance": "CORE" }, ...# Bootstrap a new client brandagent.invoke> { "agent": "brand-bootstrap", "input": { "url": "https://clientbrand.com" } }job_id: job_3mq7r · status: running · awaiting_approvalStitching raw APIs together for every SEO capability delays your agent and leaves clients waiting.
Connecting keyword research, content, indexing, and rank tracking APIs separately takes months before a single agent ships.
Generic MCP servers lack SEO-specific tools — no SERP analysis, no AI citation tracking, no link-building pipeline.
LLM calls without durable job infrastructure crash on restarts, with no approval gates or audit trail for clients.
The fix
CitationBench's hosted MCP server at mcp.citationbench.com/mcp exposes ~35 tools across research, production, distribution, and agents namespaces — callable today from Claude Code, Cursor, Windsurf, and custom MCP clients with no self-hosting required.
Five feature areas that cover the full SEO and GEO workflow — all accessible as MCP tools or REST endpoints from your agent.
MCP Infrastructure
Connect once at mcp.citationbench.com/mcp and get ~35 namespaced tools — research.*, produce.*, indexing.*, link_building.*, agent.* — no self-hosting, no glue code.
Agent Orchestration
POST /v1/agent/invoke or call agent.invoke from your MCP client to run named agents with structured input, real-time SSE streaming, and approval-gated pauses — not fire-and-forget.
Research Tools
Call research.keyword.research from any MCP client to discover, cluster, and label keyword sets with 2D intent × relevance scoring. Results persist for downstream production steps.
Job Reliability
Every agent job is backed by Cyclonic workers — resumable, cancellable, and streamable. Cancel via agent.invocations.cancel, resume via agent.invocations.continue. Nothing is lost on failure.
Agency Architecture
Scope all MCP tool calls per client via the X-Workspace-Id header. One master key manages N client workspaces with full data isolation — rank tracking, content, and campaigns all separated.
How it works
An `sk_test_*` key lands in your dashboard instantly. No demo gate — start calling real endpoints with shape-complete responses.
`claude mcp add citationbench https://mcp.citationbench.com/mcp` from Claude Code, Cursor, or any MCP client. Or `curl` against `api.citationbench.com/v1/*` directly.
Every tool returns shape-complete demo data without auth, so your agent works before the user signs up. Tools across research, production, indexing, link-building, and agents.
Add `X-Workspace-Id: ws_***` to scope per client. Switch to a live `sk_live_*` key when you're ready. Same API surface, same SDK.
Why CitationBench
Hosted at mcp.citationbench.com/mcp. Works with Claude Code, Cursor, Claude Desktop, Windsurf, and ChatGPT Apps. No self-hosting required.
Every long-running call runs on Cyclonic workers — survives restarts, cancellable, resumable, streams via SSE. Production-safe by default.
One master API key, N client workspaces. Switch with a single `X-Workspace-Id` header. Per-client data isolation, bulk ops across all of them.
Every endpoint responds in shape-complete demo mode without auth, so you can build the agent before the user signs up.
FAQ
Connect to CitationBench's hosted MCP server and get ~35 SEO tools in your agent without writing integration code.
View MCP Docs