Skip to content
GEO
Start tracking

Integrations · MCP

Your IDE already knows your visibility

Six tools, one brand-scoped token, zero dashboard tabs. Ask Cursor what slipped this week and get the same numbers the board sees — because they come from the same weekly record.

  • POST /api/mcp
  • Streamable HTTP
  • Bearer per brand
  • 60 req/min per token

Short answer

What can agents actually do?

What does the MCP server expose?

Six brand-scoped tools: current and previous citation snapshots with per-engine deltas, tracked prompts with stance per engine, source domains with opportunities, the pending action queue with draft copy — plus two scoped writes, mark-done which locks the baseline and dismiss which clears the queue. Reads match the dashboard exactly because they read the same tables.

How is it isolated and limited?

Each token is SHA-256 hashed at rest and bound to one brand — it can never see another workspace. The ceiling is sixty requests per minute per token, enforced per server instance. That last clause is stated deliberately: it is a per-instance ceiling, not a global quota, and we would rather state the constraint than imply a guarantee.

Why MCP instead of another REST export?

Because agents already speak it. Cursor, Claude Desktop, Windsurf and Claude Code call Streamable HTTP tools natively, so visibility joins the build loop — pitch slides, fix branches, reports — instead of living in exports nobody opens. CSV and JSON exports still exist for everything else.

The six tools

Same record, agent-shaped

Brand isolation is structural: the brand ID comes from the authenticated token, never from a parameter an agent could confuse.

POST https://geo.foryourreach.com/api/mcpJSON-RPC 2.0 · Streamable HTTP
  • get_citation_reportCurrent and previous snapshot, per-engine breakdown, delta.
  • list_promptsTracked buyer queries with stance, sentiment and position per engine.
  • list_sourcesSource domains, brand presence, competitor presence, opportunities.
  • list_actionsPending remediation queue with draft copy and priority.
  • mark_action_doneLocks the current snapshot as the impact baseline.
  • dismiss_actionRemoves an action from the queue.

Authorization: Bearer <token> · CORS open for browser agents · mark-done locks impactBefore from latest snapshot

6
Brand-scoped tools

4 reads + 2 writes

60
Req / min / token

Per-instance ceiling

SHA-256
Token storage

Hashed at rest

1
Brand per token

Isolation structural

Connect in minutes

Four clients, same config shape

Create a token in MCP Access settings, paste it once, ask what changed.

Cursor

Paste the Streamable HTTP URL plus token into MCP settings. Ask what changed this week in the composer.

Claude Desktop

Add the server to claude_desktop_config.json. Ask for citation reports per engine without leaving chat.

Windsurf & VS Code agents

Any Streamable HTTP client works. Scripts can poll weekly snapshots after the Monday cycle.

Your own scripts

POST JSON-RPC 2.0 with Authorization: Bearer. Parse citation reports into pitch decks and board packs.

Cursor / Claude Desktop config

{
  "mcpServers": {
    "geo-visibility": {
      "url": "https://geo.foryourreach.com/api/mcp",
      "transport": "streamable-http",
      "headers": {
        "Authorization": "Bearer <brand-token>"
      }
    }
  }
}

Ask the agent

"What dropped this week, and which fix is fastest?"
"Show Perplexity source gaps with prompts-fed counts."
"Lock the baseline — mark the llms.txt action done."
Create a token in MCP Access settings

Integration questions

Asked by platform engineers

Quoted verbatim via FAQPage markup.

Claude + MCP playbook

What can our agents read through MCP?

Six brand-scoped tools over Streamable HTTP at /api/mcp: citation reports, prompts with stance per engine, sources with opportunities, the pending action queue, plus mark-done and dismiss. Authentication is a per-brand bearer token, SHA-256 hashed at rest, at sixty requests per minute per token.

Which clients work with the MCP server?

Any Streamable HTTP client: Cursor, Claude Desktop, Windsurf, VS Code agents and your own scripts. Create a token in MCP Access settings, paste it into the client config, and ask what changed this week — the numbers match the dashboard exactly.

Is MCP access read-only?

Mostly reads, with two scoped writes: marking an action done locks the current snapshot as its baseline, and dismissing removes it from the queue. Both are brand-isolated — a token can never see another brand.

Agents welcome

Pull this week into your IDE

Reports, prompts, sources and queue — same numbers as the dashboard, callable from Cursor tonight. Fourteen days free.

Start trackingBook a 20-minute call