CitedOS ships a workspace-scoped MCP server so your own agent — Claude Desktop, an IDE assistant, anything that speaks the Model Context Protocol — can read your visibility and source-gap data and add prompts, without touching the dashboard.
The tools
| Tool | What it does | Access |
|---|---|---|
list_brands | List the brands in your workspace, with their ids. | Read-only |
get_visibility | A brand's visibility — the share of ChatGPT/Gemini/Claude answers that mention it — with 95% confidence ranges and a per-engine breakdown (ChatGPT tagged as a proxy surface). | Read-only |
get_source_gap | The source-gap analysis: which on-page traits separate cited winners from your pages, whether you're present among cited domains, and a prioritized playbook. | Read-only |
list_prompts | List the buyer-intent prompts tracked for a brand. | Read-only |
add_prompt | Add a buyer-intent prompt to a brand's set; measured on the next cycle. The one write tool. | Write |
Authentication
The server authenticates with a CitedOS API key (mint one in Settings → API keys). The key maps to a single workspace, so every tool call is scoped to your data — the same authorization boundary the web app enforces, applied to non-browser clients. Set the key as an environment variable on the server. The same key also authenticates the Agent-traffic log-ingest endpoint — one key, both integrations.
Example
CITEDOS_API_KEY=ck_… DATABASE_URL=… npm run mcp
# Then, from your MCP client (e.g. Claude Desktop), the agent can call:
# list_brands() → ids for your brands
# get_visibility(brandId: "br_…") → metrics + 95% confidence ranges
# get_source_gap(brandId: "br_…") → missing sources + playbook
# add_prompt(brandId: "br_…", text: "best CI/CD tool for startups")