curl -fsSL https://get.mbm.mn/install | bash

mbm                    # Launch interactive TUI mbm run "Hello"       # Send a message mbm --help            # Show all commands

mbm license apply MBM-XXXX-XXXX-XXXX-XXXX

mbm license activate /path/to/license.license

$ mbm license activate ./license.license
 License activated successfully.

   Device ID:  a1b2c3d4e5f6a7b8
   Tier:       pro
   Expires:    2029-12-31

mbm providers login # Select "mbm" → paste your API key (mp-abc123...)

mbm license status

  Device ID:    a1b2c3d4e5f6a7b8
  License:      Activated
  Tier:         pro
  Expires:      2029-12-31T00:00:00.000Z
  Max Devices:  1
  Priority Queue: enabled
  API Access:     enabled

mbm providers login

{
  "provider": {
    "openai": {
      "apiKey": "sk-proj-..."
    },
    "anthropic": {
      "apiKey": "sk-ant-..."
    }
  }
}

mbm models                     # List all available models mbm models openai              # Filter by provider

curl -fsSL https://ollama.com/install.sh | sh ollama pull llama3.2

{
  "provider": {
    "ollama": {
      "baseURL": "https://ollama2.mbm.mn/v1"
    }
  },
  "model": {
    "ollama": "qwen3.6:latest"
  }
}

export OLLAMA_HOST=http://your-host:11434 mbm

mbm --safe                # Launch in SAFE mode mbm run --safe "Review the auth module"

{
  "permission": {
    "bash": "ask",
    "edit": "ask",
    "read": "allow",
    "webfetch": "allow",
    "mbm-rag_*": "allow"
  }
}

mbm rag setup              # Install Docker + pgvector mbm rag enable             # Enable auto-sync on git changes mbm rag index .            # Index current directory

mbm rag search "how does auth work" mbm rag status             # View KB statistics mbm rag optimize           # Remove duplicates & stale entries

mbm mcp add               # Add an MCP server
mbm mcp list              # List configured servers
mbm mcp auth <name>       # OAuth authenticate
mbm mcp debug <name>      # Test connection

mbm session list          # List all sessions mbm session share         # Generate share link mbm session compact       # Summarize to reduce context mbm session export        # Export to file

mbm agent create          # Interactive agent builder mbm agent list            # List your agents

mbm stats                # Token usage and cost summary

# Run with a specific model (provider/model format) mbm run -m openai/gpt-4o "Refactor this to async/await" mbm run -m deepseek/deepseek-v4-pro "Explain quantum computing" mbm run -m ollama/llama3:8b "What is this error?" mbm run -m anthropic/claude-sonnet-4-5 "Review this architecture" # With thinking/reasoning enabled mbm run -m deepseek/deepseek-v4-pro:thinking --thinking "Solve this math problem"

# Pipe stdin git diff HEAD~5 | mbm run -m openai/gpt-4o "Write a PR description" cat /var/log/nginx/error.log | mbm run -m ollama/llama3:8b "Diagnose this error" # Attach files for context mbm run -m openai/gpt-4o -f src/main.ts "Find bugs in this file" mbm run -m ollama/llama3:8b -f src/*.ts "Review these files" # Batch processing for f in src/**/*.ts; do mbm run -m ollama/llama3:8b -f "$f" "Add JSDoc to untagged functions" --dangerously-skip-permissions; done

# Continue last session mbm run -c "Add error handling to the login function" mbm run --continue "Expand on the previous answer" # Resume a specific session mbm run -s sess_abc123 -m ollama/llama3:8b "Next step?" mbm run --fork -s sess_abc123 -m openai/gpt-4o "Try a different approach"

# JSON output for CI pipelines mbm run --format json -m openai/gpt-4o "Code review" > review.jsonl # Slash command mbm run --command /review -m openai/gpt-4o src/api/ # Remote server diagnostics via SSH ssh user@server "mbm run -m ollama/llama3:8b 'Server health check' --dangerously-skip-permissions"

# Auto-approve all permissions (use with caution) mbm run -m ollama/llama3:8b --dangerously-skip-permissions "Fix all bugs" # Safe mode (read-only, no writes) MBM_SAFE_MODE=1 mbm run "Analyze project structure" # Custom Ollama host OLLAMA_HOST=http://10.0.0.5:11434 mbm run -m ollama/llama3:8b "Hello" # Configure permissions in mbm.json: #   "permission": #     "external_directory":  "/proc/*": "allow" #     "bash": "allow" #     "read": "allow"

mbm providers login # Select "mbm" → paste your API key

curl https://proxy.mbm.mn/v1/chat/completions \
  -H "Authorization: Bearer mf-abc123..." \
  -H "Content-Type: application/json" \
  -d '{"model": "gpt-4o", "messages": [{"role": "user", "content": "Hello"}]}'

from openai import OpenAI

client = OpenAI(
    base_url="https://proxy.mbm.mn/v1",
    api_key="mf-abc123...",
)

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)

import OpenAI from "openai"

const client = new OpenAI({
  baseURL: "https://proxy.mbm.mn/v1",
  apiKey: "mf-abc123...",
})

const response = await client.chat.completions.create({
  model: "gpt-4o",
  messages: [{ role: "user", content: "Hello" }],
})
console.log(response.choices[0].message.content)

{
  "model": {
    "default": "openai/gpt-4o",
    "openai": "gpt-4o",
    "ollama": "llama3.2:latest"
  },
  "provider": {
    "openai": {
      "apiKey": "sk-proj-..."
    },
    "ollama": {
      "baseURL": "http://localhost:11434/v1"
    }
  },
  "permission": {
    "bash": "ask",
    "edit": "ask",
    "read": "allow",
    "webfetch": "allow"
  }
}

{
  "$schema": "https://get.mbm.mn/config.json",
  "provider": {
    "ollama": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Ollama (Local)",
      "options": {
        "baseURL": "http://localhost:11434/v1"
      },
      "models": {
        "qwen3-coder": {
          "name": "qwen3-coder:latest",
          "options": {
            "num_ctx": 262144,
            "stream": false,
            "tools": []
          }
        }
      }
    },
    "openai": {
      "apiKey": "sk-proj-..."
    }
  },
  "mcp": {
    "mbm-rag": {
      "type": "local",
      "command": [
        "python3",
        "~/.mbm/bin/rag/rag_mcp_server.py"
      ]
    }
  },
  "lsp": true,
  "rag_inject": true,
  "permission": {
    "bash": "ask",
    "edit": "ask",
    "read": "allow",
    "webfetch": "allow",
    "mbm-rag_*": "allow"
  }
}

mbm web                # starts on 127.0.0.1:8027 by default

{
  "server": {
    "port": 8027,
    "hostname": "127.0.0.1"
  }
}

export MBM_WEB_UI_BASE=/mbm mbm web

location = /mbm   { return 301 /mbm/; }
location ^~ /mbm/ {
  proxy_pass         http://127.0.0.1:8027/;   # trailing "/" strips /mbm
  proxy_buffering    off;                       # keeps SSE streams live
}

export MBM_SERVER_USERNAME=admin
export MBM_SERVER_PASSWORD='correct-horse-battery-staple'
mbm web

Manager (web)                     Developer (CLI)
─────────────                     ────────────────

1. Open http://<ip>/mbm
   → create a session in the project
2. Describe the requirements
   → plan agent researches & drafts
   a plan (read-only, safe)
3. Hand over the session
   (copy the session id / title)    ───►
                                            4. mbm attach http://localhost:8027 --session <id>
                                               (or run mbm in the project and open the session)
                                            5. Switch to build mode: /agent build
                                            6. Developer approves & the plan is implemented
                                               with full write + shell access
5. Watch progress live on the web ◄─────     (both sides stream the same session via SSE)

# Attach the terminal UI to the running web server (realtime sync)
mbm attach http://localhost:8027 --session sess_abc123

# Or continue the latest session
mbm attach http://localhost:8027 --continue

# Switch to the build agent and start implementing
/agent build

{
  "share_base_url": "http://localhost:8027"
}

export MBM_WEB_URL=http://localhost:8027

# On your laptop
ssh -N -L 8027:127.0.0.1:8027 user@<server-ip>

# Now open on your laptop
http://localhost:8027