Connect the hosted CitationBench MCP server to Claude Code in 60 seconds with one claude mcp add command, then run agentic keyword research, content, and link-building tools.

Connect the CitationBench MCP server to Claude Code in 60 seconds.

Want to try it before you sign up? Skip the prerequisites and follow the demo-mode setup — six read-only research tools work without an API key.

Prerequisites

Claude Code installed (see claude.com/code)
A CitationBench API key (sk_live_* or sk_test_*) — create one in the dashboard
A workspace ID (or use an agency master key)

Install

From any terminal:

claude mcp add citationbench https://mcp.citationbench.com/mcp \
  --header "Authorization: Bearer $CITATIONBENCH_API_KEY" \
  --header "X-Workspace-Id: $WORKSPACE_ID"

Verify:

claude mcp list

You should see citationbench with status connected.

First tool call

Open Claude Code in any project:

claude

Then ask it to use a CitationBench tool. Natural language is fine — Claude picks the right tool.

> Research keywords for project management software,
  focused on PROBLEM_SOLUTION intent.

Claude will call research.keywords and render the result inline. You'll see the request, the response, and the credits charged.

Workspace scoping

You have three options for picking a workspace.

1. Per-connection (recommended for solo workflows)

Set X-Workspace-Id at connection time (the install command above). Every tool call uses that workspace.

2. Per-call (for agency master keys)

Don't set X-Workspace-Id at connection time. Each tool will accept a workspace parameter:

> Run a rank check on the "acme" workspace.

Claude passes workspace: "ws_acme" to distribute.track_rank.

3. Bulk (for cross-workspace ops)

> Check ranks across every workspace, alert on drops of 5+.

Claude uses workspaces.bulk_action with workspaces: "all".

Multiple environments

Use different keys for dev and prod:

# Add prod
claude mcp add citationbench-prod https://mcp.citationbench.com/mcp \
  --header "Authorization: Bearer sk_live_***"

# Add sandbox
claude mcp add citationbench-test https://mcp.citationbench.com/mcp \
  --header "Authorization: Bearer sk_test_***"

In Claude Code, tools will be exposed as citationbench-prod__research_keywords, citationbench-test__research_keywords, etc.

For destructive tools (distribute.send_outreach_email, produce.publish, distribute.gsc_index), Claude Code will surface the approval prompt before each call. You can also set workspace-level approval policies in the dashboard so the API itself holds the job until a human approves — even if Claude says go.

Debugging

If a tool call fails, Claude Code shows the full request and the error code. Common ones:

unauthorized — bad API key or expired
workspace_forbidden — key doesn't have access to that workspace
insufficient_credits — top up
rate_limited — per-key rate exceeded

Check live request logs at app.citationbench.com/api-keys.

Setup the CitationBench MCP server in Claude Code