@agentaeo/mcp-server
MCP (Model Context Protocol) server for AgentAEO — run AEO (Answer Engine Optimization) audits across ChatGPT, Perplexity, Claude, and Google AI.
Source: agentaeo/agentaeo-mcp-server on GitHub (local folder name: agentaeo-mcp-server).
Installation
npm install -g @agentaeo/mcp-server
npx @agentaeo/mcp-server
Configuration
Set your API key (get one at agentaeo.com/agents):
export AGENTAEO_API_KEY=your_api_key_here
Claude Desktop
Config file on macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
After publishing to npm (recommended)
{
"mcpServers": {
"agentaeo": {
"command": "npx",
"args": ["-y", "@agentaeo/mcp-server"],
"env": {
"AGENTAEO_API_KEY": "your_api_key_here"
}
}
}
}
-y skips npx’s “Ok to proceed?” prompt (important for GUI-launched processes).
Local dev (before publish)
Prefer node + dist/index.js — npx + a local .tgz often breaks in Claude Desktop (npm path bugs, “tarball corrupted”, or Permission denied when the shell mishandles the archive). Avoid .tgz in the MCP config unless you’ve verified it on your machine.
- In the repo:
npm install and npm run build (must produce dist/index.js).
- Use your real AgentAEO key in
AGENTAEO_API_KEY (not the literal text your_real_key_here).
Recommended — node + absolute path to dist/index.js:
{
"mcpServers": {
"agentaeo": {
"command": "node",
"args": ["/Users/aashishn/agents/agentaeo-mcp-server/dist/index.js"],
"env": {
"AGENTAEO_API_KEY": "paste_your_agentaeo_live_key_here"
}
}
}
}
If node is not on Claude’s PATH (e.g. only nvm), use the full path to node, e.g. "/Users/aashishn/.nvm/versions/node/v22.16.0/bin/node" as command and the same args for dist/index.js.
Optional — tarball via npx (only if node works from Terminal but you prefer npx): some setups need file: prefix — ask in Terminal first:
npx -y file:/Users/aashishn/agents/agentaeo-mcp-server/agentaeo-mcp-server-0.1.0.tgz — if that fails, stick to node + dist.
After editing the config, fully quit Claude Desktop (Cmd+Q) and reopen. Check Settings → Developer → Local MCP servers for a green status.
Tools
run_aeo_audit | Run an AEO audit for a URL. Free: 8 queries. Paid: 40 queries with full blueprint. |
check_aeo_audit_status | Poll audit status and retrieve results. Use the auditId from run_aeo_audit. |
generate_aeo_content_suite | Generate Content Suite ZIP path (HTML + JSON-LD + llms.txt) for a completed audit. Uses the same AGENTAEO_API_KEY as the other tools (no shell). Admin testing: adminContentBypass=true + admin/allowlisted key. Production: pass orderId from aeo_content_orders after payment. Can take 10–25+ minutes — some clients may timeout. |
Why your agent can’t curl with $AGENTAEO_API_KEY
Sandbox VMs (e.g. Cowork) do not load claude_desktop_config.json and do not inherit your laptop’s shell export. Only processes started with that env (e.g. this MCP server) have the key. Use generate_aeo_content_suite from Claude Desktop with MCP enabled, or paste the key into the sandbox’s own secrets/env if the product supports it.
Example
Ask Claude: "Run a free AEO audit for agentaeo.com with keyword 'AI search visibility'"
Then: "Check the status of that audit" (Claude will use the returned auditId)
Requirements
Development & publishing
Run all npm commands from the repo root (the folder that contains package.json), e.g.:
cd ~/agents/agentaeo-mcp-server
npm install
npm run build
npm pack
Publish (after npm login and scope access for @agentaeo):
npm publish --access public
GitHub vs npm: The GitHub repo can stay private; the npm package @agentaeo/mcp-server is usually public so npx @agentaeo/mcp-server works for everyone. Those are independent settings.
License
MIT