Cross-product tutorial

Unified AI CLI Access for Teams

Consolidate the AI CLI configs, MCP servers, and model keys scattered across every engineer's machine into one central config: Switch handles MCP and sync, Lurus API handles models and billing.

Products involved

Switch (desktop tool management) · Lurus API (unified gateway) · MCP servers (Kova / GitHub / PostgreSQL, etc.). This tutorial only references capabilities already documented for each product.

What you'll get

Before (every person on their own)	After (unified with Switch)
Everyone hand-writes `mcp_servers.json`, with inconsistent tool versions	One central `mcp.yaml`, with `visible_to` delivering on demand
Each CLI gets its own set of Provider keys	All traffic goes through Lurus API — one key, one bill
Model costs are invisible	Switch cost dashboard aggregates by tool / model

Step 1

Install Switch and connect the gateway

Install Switch following the installation guide (macOS / Windows / Linux).
In Settings, enter your Lurus API Key (how to get one) so every CLI calls models uniformly through https://hub.lurus.cn/v1 — one key, one bill.
Verify the local proxy is running (default port 19090):
bash
```
curl http://localhost:19090/v1/models
```
1

Step 2

Write one central MCP config

Switch manages all MCP servers from a single ~/.lurus-switch/mcp.yaml, and visible_to decides which ones each CLI sees.

yaml

servers:
  github:
    command: npx
    args: [-y, '@modelcontextprotocol/server-github']
    env:
      GITHUB_TOKEN: ${GITHUB_TOKEN}
    visible_to: [claude-code, codex]

  postgres:
    command: docker
    args: [run, -i, --rm, mcp/postgres, 'postgres://localhost/dev']
    visible_to: [claude-code]

  kova:                       # Kova Agent 作为工具暴露
    url: http://localhost:3333
    type: http
    visible_to: [claude-code, codex, gemini]

visible_to drives on-demand delivery

When you switch to any CLI, Switch dynamically generates that CLI's mcp_servers.json based on visible_to, so each tool only sees the servers assigned to it. See the integration catalog for the list of connectable servers, and MCP Servers for management details.

Step 3

See the costs, then sync to the team

Open Switch's cost monitoring to view token consumption by tool / model — since all calls go through the same Lurus API Key, the bill is unified.
Once the config is confirmed correct, use team sync to distribute this mcp.yaml to the team, so new members are ready out of the box with a consistent version.

What you can add next

Connect MemX memory to your CLI (the memory_search / memory_add tools) so the agent remembers project conventions; or connect Lumen for call tracing and cost alerts.

Next steps

Switch MCP Servers Integrations & MCP catalog Memory Agent tutorial