@site-spy/mcp-server

@site-spy/mcp-server

Let an AI agent watch web pages for you. This MCP server connects your assistant to Site Spy: it can start monitoring a URL, tell you what changed between two points in time, and fetch or search the web on demand. Full docs at docs.sitespy.app.

Install

Get an API key from Settings → API. Every plan, including the free one, can read and write — you are bounded by watch count and daily requests, not by which tools you may call.

If your client has a CLI, this is the whole setup:

# Claude Code
claude mcp add site-spy --env SITE_SPY_API_KEY=your-key -- npx -y @site-spy/mcp-server

# Codex CLI
codex mcp add site-spy --env SITE_SPY_API_KEY=your-key -- npx -y @site-spy/mcp-server

# Gemini CLI  (the -- matters: without it, -y is parsed as a gemini flag)
gemini mcp add -s user -e SITE_SPY_API_KEY=your-key site-spy npx -- -y @site-spy/mcp-server

# OpenClaw
openclaw mcp add site-spy --command npx --arg -y --arg @site-spy/mcp-server \
  --env SITE_SPY_API_KEY=your-key

# VS Code
code --add-mcp '{"name":"site-spy","command":"npx","args":["-y","@site-spy/mcp-server"],"env":{"SITE_SPY_API_KEY":"your-key"}}'

Otherwise add this to your client's config file:

{
  "mcpServers": {
    "site-spy": {
      "command": "npx",
      "args": ["-y", "@site-spy/mcp-server"],
      "env": {
        "SITE_SPY_API_KEY": "your-key"
      }
    }
  }
}

Client	Config file	Shape
Claude Code	.mcp.json (project) or ~/.claude.json	as above
Claude Desktop	~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json	as above
Cursor	.cursor/mcp.json or ~/.cursor/mcp.json	as above
Windsurf	~/.codeium/windsurf/mcp_config.json	as above
Gemini CLI	.gemini/settings.json or ~/.gemini/settings.json	as above
Codex CLI	~/.codex/config.toml	TOML, see below
OpenCode	opencode.json or ~/.config/opencode/opencode.json	differs, see below
OpenClaw	~/.openclaw/openclaw.json	differs, see below
VS Code	.vscode/mcp.json	differs, see below
Zed	run the zed: open settings file action	differs, see below

Restart the client afterwards; the tools appear on their own.

Clients whose config shape differs

Codex CLI — TOML, and the key is mcp_servers. The same file is read by the Codex CLI, the ChatGPT desktop app, and the IDE extension.

[mcp_servers.site-spy]
command = "npx"
args = ["-y", "@site-spy/mcp-server"]

[mcp_servers.site-spy.env]
SITE_SPY_API_KEY = "your-key"

OpenCode — key is mcp, command is one array, and the env key is environment.

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "site-spy": {
      "type": "local",
      "command": ["npx", "-y", "@site-spy/mcp-server"],
      "enabled": true,
      "environment": { "SITE_SPY_API_KEY": "your-key" }
    }
  }
}

OpenClaw — servers nest under mcp.servers.

{
  "mcp": {
    "servers": {
      "site-spy": {
        "command": "npx",
        "args": ["-y", "@site-spy/mcp-server"],
        "env": { "SITE_SPY_API_KEY": "your-key" },
        "enabled": true
      }
    }
  }
}

VS Code — key is servers, and inputs keeps the key out of the file.

{
  "inputs": [
    {
      "type": "promptString",
      "id": "site-spy-key",
      "description": "Site Spy API key",
      "password": true
    }
  ],
  "servers": {
    "site-spy": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@site-spy/mcp-server"],
      "env": { "SITE_SPY_API_KEY": "${input:site-spy-key}" }
    }
  }
}

Zed — key is context_servers.

{
  "context_servers": {
    "site-spy": {
      "command": "npx",
      "args": ["-y", "@site-spy/mcp-server"],
      "env": { "SITE_SPY_API_KEY": "your-key" }
    }
  }
}

If you skip SITE_SPY_API_KEY, the agent asks you to authenticate interactively — but that key lives in memory only and is gone when the server restarts.

Tools

Read tools are annotated readOnlyHint, so most clients stop asking permission for them.

Tool	Does	Plan
list_watches	List monitored pages, with limit / offset	all
get_watch	Full detail for one watch	all
search_watches	Find watches by URL or title	all
get_change_history	Timestamps where a change was detected	all
get_snapshot	The page as captured at one timestamp	all
get_diff	What changed between two timestamps	all
get_rss_settings	Current feed settings and token	all
get_notifications	Current notification settings	all
auth_status	Whether the server is connected	all
create_watch	Start monitoring a URL	all
update_watch	Rename, repoint, pause, change interval, mute	all
delete_watch	Stop monitoring and drop the history	all
trigger_recheck	Check everything now instead of on schedule	all
authenticate	Supply an API key at runtime	all
generate_rss_token	Mint a feed token — invalidates existing feed URLs	all
revoke_rss_token	Kill feed access immediately	all
fetch_url	Fetch one URL, with a browser backend for JS pages	all
web_search	Find URLs to fetch, instead of guessing them	all
get_ai_summary	Plain-English summary of the latest change	Pro

get_snapshot and get_diff cap output at 20,000 characters; pass max_chars for more.

Once a token exists, feed URLs are https://sitespy.app/api/rss?token={token} for everything and …/api/rss/watch/{uuid}?token={token} for one watch.

Environment variables

Variable	Description	Default
SITE_SPY_API_KEY	API key	— (interactive auth)
SITE_SPY_API_URL	Backend API URL	https://api.sitespy.app/api/v1
SITE_SPY_AUTH_URL	URL shown when a key is needed	https://sitespy.app/dashboard

Self-hosting: the backend fetch_url endpoint can be restricted with FETCH_URL_API_ENABLED, FETCH_URL_REQUIRE_MASTER_ADMIN, or FETCH_URL_ALLOWED_API_KEYS.

Troubleshooting

• "API key is invalid or expired" — the key was revoked or regenerated. Copy a fresh one from Settings → API. The server itself is fine; there is no need to work around it.
• Tools never appear — restart the client after editing its config, and check npx runs (Node.js 20.19+).
• "Watch limit reached" — you are at your plan's cap. Delete one or upgrade.
• "Rate limit reached" — the day's write allowance is spent; the error says how long to wait. Reads do not count against it.
• fetch_url keeps timing out — pass fetch_backend explicitly. html_requests is fast and right for static pages; html_webdriver renders JavaScript; flaresolverr is for Cloudflare. auto tries all three and gives up at 30s. On failure the response carries attempts[] and cascade_summary showing what was tried.

Try it

• "Monitor https://example.com/pricing and tell me when the price changes"
• "What changed on my watches this week?"
• "Search for the current Postgres release notes and fetch the top result"