Data that only lives in one dashboard gets looked at once a month. SOV Tracker gives you two doors into the same numbers: a REST API that feeds your reporting stack, and an MCP server that puts your Share of Voice one question away in Claude, ChatGPT or another MCP client. Every MCP tool is read-only.
Endpoint
https://sovtracker.com/api/mcp (Streamable HTTP)
Authorization: Bearer avt_live_••••••••••••The same Bearer key authenticates the REST endpoints under /api/v1. Keys are shown once, on creation.
No SDK to install and no separate integration to maintain. The key you create in the dashboard is the only credential involved.
Keys are created in the dashboard under Settings. Every key starts with avt_live_ and carries a read permission; the MCP server only authorizes keys that have it. API and MCP access is part of the Pro plan and above.
Point your MCP client at the endpoint below. Claude.ai and ChatGPT connect by signing in to SOV Tracker, so there is no key to paste. Claude Code, Cursor and other clients send the same Bearer key as the REST API. Either way there is no extra install and no second password to manage.
Ask your assistant a question about your visibility and it calls the matching read-only tool, then answers from your own scan data instead of guessing. The same tools are available to your scripts through the REST API.
claude mcp add --transport http sovtracker https://sovtracker.com/api/mcp --header "Authorization: Bearer avt_live_••••••••••••"Generate the key under Settings → API in the dashboard. Pro plans and up.
Run in Claude Code, then check the connection with /mcp. Use your own key in place of the masked example.
ChatGPT: add a custom connector with the same URL and sign in to SOV Tracker; no key is needed.
Three surfaces over one dataset: tools for your assistant, endpoints for your code, and a webhook that tells your systems when fresh data landed.
Every tool in this version is read-only and scoped to the organization that owns the key. There is deliberately no scan-trigger tool: an assistant that loops cannot burn your scan quota. Reading is safe by construction, so you can hand the connection to your whole team.
Five endpoints under /api/v1 include GET /api/v1/scorecard for the canonical seven-metric weekly scorecard, GET /api/v1/sov for legacy/current scan compatibility, GET/POST /api/v1/scans, and GET /api/v1/archive. Same Bearer key, JSON responses.
The scan.completed event is delivered to your endpoint when a scan finishes, with the event name in the X-Sovtracker-Event header and an HMAC-SHA256 signature of the payload in X-Sovtracker-Signature so you can verify it. Endpoints are managed in dashboard Settings.
This is the complete list: 7 tools, all read-only, all scoped to the organization that owns the key.
| Tool | What it returns |
|---|---|
| get_metric_scorecard | Canonical weekly Measurement Contract v1.0 scorecard: SOV, Discovery, Recommendation, Owned-Site Citation Coverage, Sentiment, Prominence and Brand Accuracy, with missing states preserved. |
| get_visibility_summary | Latest completed scan: overall Share of Voice score, per-platform score and whether the brand was mentioned, plus the trend against the previous scan. |
| list_keywords | The questions your organization tracks in AI answers, the same list as the dashboard: status (active or paused), priority flag, market and language. |
| get_scan_results | Row-level results for a scan: platform, question, mention, SOV score, sentiment and mention position, filterable by platform or question. |
| get_answer_archive | Dated raw AI answers from your weekly measurements, the same answers as the dashboard archive, so you can read what a platform actually said on a given day. Long answers are truncated and paginated with a cursor. |
| get_business_card | Brand Facts card status: trust score, per-field-group freshness and the active-business attestation date. Read-only, like every other tool here. |
| get_ai_traffic | Human visits associated with AI tools over the last N days, reported in two separate signal classes: direct AI referral and AI-tagged. They are never merged into a single referral number. |
Read-only by design: no tool in this version writes data or starts a scan, so a looping assistant cannot spend your scan quota.
60 requests per minute per key on the MCP server, and 60 per minute per organization on GET /api/v1/archive. Over the limit you get a 429, never a silent failure.
Both share one gate: the Pro plan and above. A key belonging to a lower plan is rejected before it reaches any data, and the same key works for the REST endpoints and the MCP server, so there is nothing extra to buy or wire up once you are on Pro.
No. This version of the MCP server exposes read tools only, and no tool can trigger a scan. That is a deliberate cost guard. Scans stay under your control: you start them from the dashboard, or from your own code with POST /api/v1/scans, which allows at most 30 scan starts per hour.
The MCP server allows 60 requests per minute per key. GET /api/v1/archive allows 60 requests per minute per organization, shared by all of its keys. Starting scans is limited separately to 30 per hour per organization. Going over a limit returns 429 rather than failing silently, so your integration can back off and retry.
scan.completed is the event that fires on its own, every time a scan finishes, with no extra configuration. It is signed with your endpoint secret so you can verify that the request really came from us. Other events depend on your alert rules; the full list and payloads are in the API reference.
Create a key in the dashboard, paste the endpoint into your MCP client, and ask your first question. API and MCP access is part of the Pro plan and above.