MCP server
Two tools that let a coding agent scan a site for AI readiness and read the fixes, without leaving the repo.
Server
POST https://mcp.sagbot.app/mcpStreamable HTTP
- Auth: a personal API token as
Authorization: Bearer sgb_…(get a token). Sign-in from the client (OAuth) isn't supported. - Stateless: every request is a
POSTanswered with plain JSON. There is no session and no server-sent stream, soGETandDELETEanswer405. - Setup for Claude Code, Cursor, Codex, Gemini CLI, VS Code and Windsurf, with your token filled in: sagbot.app/mcp.
Scans run against the public site, so deploy your fixes (or use a public preview URL) before a rescan. MCP scans show up in your Dashboard.
scan_site
Scans a site: Agent-ready and Answer-ready scores, and every failed or partial check with fix instructions written for a coding agent. A scan can take up to a minute.
| Argument | Type | Description |
|---|---|---|
domain | string, required | Domain or URL of the live site, such as example.com |
siteType | string | content (blog, docs, marketing; the default), app (SaaS, web app) or shop |
rescan | boolean | true scans again instead of reusing a recent result. Default false |
A result for the same site and type from the last 60 minutes (5 minutes if it didn't complete) is reused, for free. rescan: true runs a fresh scan at most once per site every 2 minutes; inside that window you get the site's newest stored scan, which may be of another siteType, and rescanInSeconds. Every call counts toward the 3 a minute, cached answers included.
If the call times out, the scan still finishes. Call get_latest_scan for it instead of scanning again.
get_latest_scan
Reads the newest stored scan of a site and type, complete or not, without scanning. Free, and it doesn't count toward the scan limit. Use it after a scan_site call that timed out, or to re-read the fixes.
| Argument | Type | Description |
|---|---|---|
domain | string, required | Domain or URL of the site |
siteType | string | content (default), app or shop |
With no stored scan it answers, as plain text, that there is none yet and to call scan_site.
Tool output
Both tools return the scan twice: as text for the agent to read, and as structuredContent for code. The fields match the Scan API response, in camelCase and without cost_usd. Fix texts are shortened here.
Text
SagBot scan of example.com (content), 2026-11-17T09:41:07.000Z
Agent-ready: 62/100 · Answer-ready: 71/100
Fixes, by impact (2):
1. [fail] llms.txt (agent-llms-txt)
https://example.com/llms.txt returns HTTP 404. Serve /llms.txt as plain text or Markdown: …
2. [partial] Canonical URL (seo-canonical)
The homepage of example.com has no canonical link. Add exactly one absolute <link rel="canonical"> …
Couldn't verify: MCP server card (agent-mcp-card)
Passing: 31 · Info: 4 · n/a: 3
Report: https://sagbot.app/d/example.comstructuredContent
{
"host": "example.com",
"siteType": "content",
"scannedAt": "2026-11-17T09:41:07.000Z",
"cached": false,
"complete": true,
"scores": { "agentReady": 62, "answerReady": 71 },
"fixes": [
{ "id": "agent-llms-txt", "title": "llms.txt", "status": "fail", "fix": "https://example.com/llms.txt returns HTTP 404. …" },
{ "id": "seo-canonical", "title": "Canonical URL", "status": "partial", "fix": "The homepage of example.com has no canonical link. …" }
],
"couldNotVerify": [{ "id": "agent-mcp-card", "title": "MCP server card" }],
"counts": { "pass": 31, "info": 4, "na": 3 },
"report": "https://sagbot.app/d/example.com"
}fixes lists every failed or partial check, highest impact first, with its fix text as written. rescanInSeconds and failReason appear only when they apply; report is the public report page.
Call it without a client
The server is stateless, so a single JSON-RPC request is enough; no initialize handshake is kept between calls. Send both types in Accept.
curl https://mcp.sagbot.app/mcp \
-H "Authorization: Bearer $SAGBOT_TOKEN" \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-d '{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": { "name": "get_latest_scan", "arguments": { "domain": "example.com" } }
}'For scripts and CI, the Scan API does the same over plain REST.
Limits and prices
| What | Limit or price |
|---|---|
scan_site calls | 3 a minute per account, cached answers included, shared with POST /v1/scan |
Requests to mcp.sagbot.app | 60 a minute per IP address |
| Scans of your own verified domains | Free, no cap (unless the domain redirects to another site) |
| New scans of other sites | First 3 a day (UTC) free, shared with the Scan API; then $0.06 each from your balance |
Stored results, get_latest_scan | Free |
Top up at Account → Billing. More on pricing: Getting started.
Errors
Tool errors come back as a tool result with isError: true and the reason as text:
- an unreadable domain or an unknown
siteType; Rate limit: 3 scans a minute, retry in ~60 s;- the site's results are hidden by its owner or SagBot and you aren't its verified owner (
scan_siteandget_latest_scan; not charged); - today's free scans are used and the balance can't pay $0.06: the message links to Billing.
Requests that never reach a tool get an HTTP status with a JSON-RPC error:
| Status | JSON-RPC code | When |
|---|---|---|
| 401 | -32001 | Missing, unknown or revoked token (with WWW-Authenticate: Bearer) |
| 405 | -32000 | GET or DELETE: only POST is served |
| 429 | -32000 | Over 60 requests a minute from your IP (with Retry-After: 60) |