使用说明
AI Builder MCP
An MCP server that provides deployment guidance and API usage support for the AI Builders platform. This README is for end users and focuses on installation and usage, not development.
Install
npm install -g @aibuilders/mcp-coach-server
You can also run without a global install:
npx @aibuilders/mcp-coach-server
Run
The server reads AI_BUILDER_TOKEN from your environment.
AI_BUILDER_TOKEN=your_token_here mcp-coach-server
Or with npx:
AI_BUILDER_TOKEN=your_token_here npx mcp-coach-server
Configure in MCP Clients
Example (Claude Desktop mcpServers):
{
"mcpServers": {
"ai-builder-mcp": {
"command": "mcp-coach-server",
"env": {
"AI_BUILDER_TOKEN": "your_token_here"
}
}
}
}
Deployment Guidance
The Coach provides instructions and API metadata; your AI assistant makes the deployment requests.
POST /v1/deploymentsreturns 202 Accepted after the task is queued, not after deployment succeeds.streaming_logsis null in deployment POST and status responses. Empty inline logs do not indicate a bad repository or branch.- Poll
GET /v1/deployments/{service_name}while status isqueuedordeploying. Do not submit another POST just because the task is still pending or a response timed out. - Fetch build logs through
/v1/deployments/{service_name}/logs?log_type=build&timeout=5, or runtime logs withlog_type=runtime. Logs come from Koyeb on demand; each call returns JSON after a bounded collection window. - After
HEALTHY, verify the public URL before reporting success. Use live logs to diagnoseERROR,UNHEALTHY, orDEGRADED.
Deployment guidance and OpenAPI metadata are cached for 24 hours. To fetch fresh data, call get_deployment_guide or get_api_specification with {"force_refresh": true}. When upgrading the package, restart the MCP server so the new version is loaded.
License
MIT
