Export API
Authentication
Create a team API key under Account settings → API keys and send it as a bearer token. Keys are scoped to one team. They read everything and can start evals (POST /sites/:id/evals/run); they can't change settings.
curl https://agentstats.io/api/v1/sites \
-H "Authorization: Bearer al_sk_…"
Conventions
- Base URL:
https://agentstats.io/api/v1 - CSV: add
?format=csvto any list endpoint. - Range:
rangeis one oftoday(since midnight in the project's timezone),1h,12h,24h,3d,7d,14d,30dor90d. Default7d. - Filters: overview, tools, errors and sessions also take
country(ISO code, orcloudfor AI providers' servers and CI),agent(a client name such asclaude-code) andagentType(coding,assistant,browser,human,ciorother). - Source: a project's tool calls come from its website (WebMCP), its MCP servers and its CLIs. Overview, tools, errors, sessions, events and schema take
source=weborsource=<MCP server or CLI id>to cover one of them; without it they cover the whole project. Per-tool rows always include theirsource, since two sources can have a tool of the same name. - Pagination: cursor-based with
beforeandlimit. Keep requesting withbefore=<last id>until a page has fewer rows thanlimit.
Endpoints
Account
GET /api/v1
Who the key belongs to, plus the list of endpoints.
Sites
GET /api/v1/sites
Every project in the team, with its id. Use that id in the endpoints below.
Overview
GET /api/v1/sites/{id}/overview?range=30d
Totals, trend and top tools for the range.
Tools
GET /api/v1/sites/{id}/tools?range=7d
GET /api/v1/sites/{id}/tools/{tool}
Per-tool call counts, error rates and latency. The second form returns one tool in detail.
Errors
GET /api/v1/sites/{id}/errors
Failed tool calls, grouped.
Sessions
GET /api/v1/sites/{id}/sessions?limit=100&before={iso}
GET /api/v1/sites/{id}/sessions/{session_id}
Agent sessions, newest first. The second form returns every event in one session.
Events
GET /api/v1/sites/{id}/events?limit=1000&before={id}&since={iso}&until={iso}
The bulk-export endpoint: raw events, newest first. Optional filters: type (e.g. tool_call), tool, status (e.g. error) and session.
Schema
GET /api/v1/sites/{id}/schema
The current tool surface: names, descriptions, input schemas and annotations.
Alerts
GET /api/v1/sites/{id}/alerts?limit=100
Schema changes, new and removed tools, and error spikes.
Visitors
GET /api/v1/sites/{id}/visitors?range=30d
GET /api/v1/sites/{id}/visitors?range=30d&ref=ChatGPT&page=/pricing
GET /api/v1/sites/{id}/visitors?range=30d&format=csv&breakdown=sources
The people on the website, from Visitors: totals with the previous period, the series, sources, UTM tags, top, entry and exit pages, countries, devices and goals. Filter with the dashboard's parameters: page, ref (source, (none) for direct), utm_medium, utm_source, utm_campaign, country, device, browser, os and goal. With ?format=csv you get the series, or one breakdown with breakdown=sources, utm_medium, utm_source, utm_campaign, pages, entry_pages, exit_pages, countries, devices, browsers, os or goals.
Traffic
GET /api/v1/sites/{id}/traffic?range=30d
Human visitors vs. agent sessions. With ?format=csv you get the daily series. Human numbers come from Visitors once the site sends pageviews, else from an analytics tool connected before imports were paused. searchClicks adds up Google Search Console and Bing Webmaster Tools, and searchEngines splits them (Search engines & SEO).
Days and chart buckets everywhere in the API start at midnight in the project's timezone (Project settings → General); timestamps are ISO 8601 in UTC.
AI bots
GET /api/v1/sites/{id}/bots?range=7d&scope=ai
GET /api/v1/sites/{id}/bots/hits?limit=200&before={id}
A summary of AI crawler and agent traffic. scope is ai (default) or all. The hits form returns individual requests and accepts product, agent, kind (bot or referral), verified and format=csv.
AI visibility
GET /api/v1/sites/{id}/visibility
GET /api/v1/sites/{id}/visibility?run={run_id}
What ChatGPT, Claude, Gemini, Perplexity and Google AI Mode say about the product: the profile, its runs, and the results of the latest finished run (or the one run names).