Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@AI Sales MCP Server Show me the top 5 products by revenue this month."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Exposes ERP data to Cursor / Claude via MCP.
settings.py → reads .env (backend URL, API key, user id)
erp_api.py → get_erp_data("/sales") calls Express
server.py → @mcp.tool functions + mcp.run()Cursor / Claude → server.py (tool) → erp_api.py → Express → PostgresRelated MCP server: BOS MCP Server
MCP talks to the Express backend. Start backend (and DB) first, then MCP.
# From repo root — ensure Postgres is running and DATABASE_URL is set
cd backend
cp ../.env.example ../.env # or use backend/.env
npm install
npx prisma migrate deploy
npx tsx prisma/seed.ts
npm run devBackend: http://localhost:4000
cd mcp_server
uv sync
cp .env.example .envEdit .env (see Environment variables below), then use one of the run modes in the sections that follow.
Copy from .env.example:
BACKEND_URL=http://localhost:4000
INTERNAL_API_KEY=dev-internal-key-change-me
MCP_ACTING_USER_ID=
MCP_TRANSPORT=stdioVariable | Required | Where to get the value |
| Yes | Express server URL. Local default: |
| Yes | Must match the backend’s |
| Yes | ERP user id (cuid) used for RBAC. Get it after seeding: login as |
| No | Documented default is |
Example after seeding (id will differ on your machine):
MCP_ACTING_USER_ID=cmrxavwxt008quumiiag90vui # e.g. admin@acme.comcd mcp_server
uv run fastmcp inspect server.pyJSON report:
uv run fastmcp inspect server.py --format mcp
# or write to a file:
uv run fastmcp inspect server.py --format mcp -o inspect.jsonStarts the server with the MCP Inspector for trying tools in the browser:
cd mcp_server
uv run fastmcp dev inspector server.pyOptional ports:
uv run fastmcp dev inspector server.py --ui-port 6274 --server-port 6277Ensure .env is filled and the Express backend is running before calling tools.
uv run python server.py
# or
uv run fastmcp run server.py --transport stdioInstall this server into Claude Desktop (writes Claude’s MCP config):
cd mcp_server
uv run fastmcp install claude-desktop server.py \
--name ai-sales-erp \
--env-file .envOr pass env vars explicitly:
uv run fastmcp install claude-desktop server.py \
--name ai-sales-erp \
--env BACKEND_URL=http://localhost:4000 \
--env INTERNAL_API_KEY=dev-internal-key-change-me \
--env MCP_ACTING_USER_ID=YOUR_USER_IDThen restart Claude Desktop. Claude launches the MCP process via stdio; keep the Express backend running on BACKEND_URL.
Config file (macOS): ~/Library/Application Support/Claude/claude_desktop_config.json
cd mcp_server
uv run fastmcp install cursor server.py \
--name ai-sales-erp \
--env-file .envmcp.json{
"mcpServers": {
"ai-sales-erp": {
"command": "/Users/pratik/Work/ai-sales/mcp_server/.venv/bin/python",
"args": ["server.py"],
"cwd": "/Users/pratik/Work/ai-sales/mcp_server",
"env": {
"BACKEND_URL": "http://localhost:4000",
"INTERNAL_API_KEY": "dev-internal-key-change-me",
"MCP_ACTING_USER_ID": "YOUR_USER_ID"
}
}
}
}Update paths for your machine. Restart Cursor / reload MCP after changes.
Local Claude/Cursor installs use stdio (client spawns server.py). For a remote MCP, run the server as an HTTP process and point clients at its URL.
cd mcp_server
# Backend must be reachable from this host (set BACKEND_URL in .env)
uv run fastmcp run server.py --transport http --host 0.0.0.0 --port 8000Default path is /mcp/, so the endpoint is:
http://<host>:8000/mcp/Use your public hostname / reverse proxy URL in production.
Claude Desktop’s config file prefers stdio. Bridge HTTP with mcp-remote:
{
"mcpServers": {
"ai-sales-erp": {
"command": "npx",
"args": ["-y", "mcp-remote", "http://YOUR_HOST:8000/mcp/"]
}
}
}Or add a custom connector in Claude: Settings → Connectors → Add custom connector → paste https://YOUR_HOST/mcp/ (for internet-reachable servers; OAuth if required).
In Cursor MCP settings / mcp.json, use a URL entry (Streamable HTTP):
{
"mcpServers": {
"ai-sales-erp": {
"url": "http://YOUR_HOST:8000/mcp/"
}
}
}If your Cursor build only supports stdio, use the same npx mcp-remote ... bridge as Claude Desktop.
uv run fastmcp list http://YOUR_HOST:8000/mcp/
uv run fastmcp inspect http://YOUR_HOST:8000/mcp/server.py)Tool | What it does |
| Find employees |
| List sales with filters |
| Find customers |
| Dashboard metrics |
| Top customers by revenue |
| Top products by revenue |
Open server.py and copy this pattern:
@mcp.tool
async def my_new_tool(name: str) -> str:
"""Short description for the AI."""
data = await get_erp_data("/some/path", {"search": name})
return to_json(data)Restart the MCP client (or reload MCP) so it picks up the new tool.
This server cannot be installed
Unclaimed servers have limited discoverability.
If you are the server author, to access and configure the admin panel.
Connect e-commerce and marketing data to AI assistants via MCP.
100+ MCP tools for AI agents: content metadata, trade intelligence, business-expertise analysis.
Self-hosted MCP gateway: turn any API, database or MCP server into AI connectors — no code.
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/pratik-eternal/ai-sales-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server