Scan your site and fix it without leaving your editor
The SagBot MCP server lets your AI coding tool scan a live site, read every failed check with its fix, and rescan after you deploy.
Works with Claude Code, Cursor, OpenAI Codex CLI, Gemini CLI, VS Code (Copilot agent mode) and Windsurf.
Get startedHow it works
Connect
Create a token and add the server to your tool. One command or one config entry.
Ask for a scan
Tell your agent to scan your site. It gets the scores and a copy-paste fix for every failed check.
Fix, deploy, rescan
Your agent edits the code, you review and deploy, then it rescans to check the score went up.
Get started
Create a free account, make a token and paste it into your tool. It takes about two minutes.
Set up your tool
Pick your tool. The snippets show a placeholder until you create a token.
Claude Code
Run this in a terminal. It adds SagBot for all your projects:
claude mcp add --transport http --scope user sagbot https://mcp.sagbot.app/mcp --header "Authorization: Bearer sgb_…"Check: claude mcp list should show sagbot ✔ Connected. In Claude Code, /mcp shows it too.
Cursor
Add this to ~/.cursor/mcp.json (all projects). If the file already has mcpServers, add sagbot inside it:
{
"mcpServers": {
"sagbot": {
"url": "https://mcp.sagbot.app/mcp",
"headers": {
"Authorization": "Bearer sgb_…"
}
}
}
}Check: sagbot shows, switched on, in Cursor’s MCP settings. If it doesn’t connect, Output → MCP Logs says why.
OpenAI Codex CLI
Add this to ~/.codex/config.toml:
[mcp_servers.sagbot]
url = "https://mcp.sagbot.app/mcp"
bearer_token_env_var = "SAGBOT_TOKEN"Codex reads the token from the SAGBOT_TOKEN environment variable. Put this line in a file outside any repository, such as ~/.config/sagbot.env (chmod 600), add source ~/.config/sagbot.env to your shell profile and open a new terminal:
export SAGBOT_TOKEN="sgb_…"Check: codex mcp list lists sagbot. In Codex, /mcp shows its tools.
Gemini CLI
Add this to ~/.gemini/settings.json. If the file already has mcpServers, add sagbot inside it:
{
"mcpServers": {
"sagbot": {
"httpUrl": "https://mcp.sagbot.app/mcp",
"headers": {
"Authorization": "Bearer sgb_…"
}
}
}
}Check: gemini mcp list shows sagbot as Connected. In Gemini CLI, /mcp lists its tools.
VS Code (Copilot agent mode)
Add this to .vscode/mcp.json in your project, or to your user mcp.json (Command Palette → MCP: Open User Configuration):
{
"inputs": [
{
"type": "promptString",
"id": "sagbot-token",
"description": "SagBot token",
"password": true
}
],
"servers": {
"sagbot": {
"type": "http",
"url": "https://mcp.sagbot.app/mcp",
"headers": {
"Authorization": "Bearer ${input:sagbot-token}"
}
}
}
}There’s no token in the file: VS Code asks for it the first time the server starts and stores it securely. Paste this when it asks:
sgb_…Check: Command Palette → MCP: List Servers shows sagbot running. In Chat, agent mode lists its tools.
Windsurf
Add this to ~/.config/devin/mcp_config.json (older Windsurf builds: ~/.codeium/windsurf/mcp_config.json). To be sure you open the right one, use Cascade’s MCP menu → Open MCP config file. If the file already has mcpServers, add sagbot inside it:
{
"mcpServers": {
"sagbot": {
"serverUrl": "https://mcp.sagbot.app/mcp",
"headers": {
"Authorization": "Bearer sgb_…"
}
}
}
}Check: after saving, press refresh in Cascade’s MCP settings. sagbot shows with 2 tools.
Two tools
scan_site
Scans a live site and returns the Agent-ready and Answer-ready scores, plus every failed or partial check with fix instructions written for a coding agent.
domain: the site to scan, for example example.com.siteType: content (default), app or shop.rescan: true runs a fresh scan instead of reusing a result under an hour old.
get_latest_scan
Reads the newest stored scan of a site without scanning and without using your limit. Use it to re-read the fixes, or after a scan_site call timed out: the scan still finishes.
domain,siteType: as in scan_site.
Prompts to try
Paste one into your agent once it is connected.
- Scan example.com with SagBot and fix what you can in this repo.
- I just deployed. Rescan example.com with SagBot and tell me what is still failing.
- What does SagBot say about the llms.txt of example.com?
Limits
- 3 scans a minute per account, however many tokens you have.
- A site can be rescanned once every 2 minutes.
- Public URLs only. SagBot fetches your site from the internet, so deploy first or use a public preview URL.
- Results are public, like any SagBot scan: anyone with the domain can open its result page.
Questions
Is the MCP server free?
Yes. You need a free SagBot account to create a token. There is no daily cap, only the limit of 3 scans a minute.
Which AI tools work?
Claude Code, Cursor, OpenAI Codex CLI, Gemini CLI, VS Code with Copilot agent mode and Windsurf. A client that can send an Authorization header to a remote MCP server can connect too, but these six are the ones we document.
Why not Claude Desktop, claude.ai or ChatGPT?
Their connectors sign in through the browser with OAuth, which SagBot does not offer yet. Tokens work in tools where you set the header yourself.
Is my token safe?
SagBot shows the token once and cannot show it again. Keep it out of git: use your user-level config, not a file in the repository. Anyone holding the token can run scans as you.
How do I revoke a token?
Open Account, then Credentials, and delete the token. The tool using it loses access.
Can it scan localhost?
No. SagBot fetches your site from the internet, so it needs a public URL. Deploy first, or use a public preview URL.
Does SagBot change my code?
No. SagBot only scans and reports. Your agent edits your code from the fixes, and you review and deploy.
Do MCP scans show up in my dashboard?
Yes. They appear in your Dashboard, marked as run via MCP.