Examples

Real-world usage patterns for mcp-setu.

Example 1: File Analysis with Filesystem Server

Analyze code files in your project:

# Set up config
cat > mcp.json << 'EOF'
{
  "ollama": {
    "model": "llama3.2:3b",
    "systemPrompt": "You are an expert code reviewer. Analyze code structure, performance, security, and best practices. Use the filesystem to read and examine files. Provide actionable feedback and suggestions for improvement."
  },
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/project"]
    }
  }
}
EOF

# Start chat
mcp-setu chat

In the chat:

❯ review the main.go file and suggest improvements

⚙ filesystem › read_file {"path": "main.go"}
↳ read_file [file content...]

setu: I've reviewed main.go. Here are the suggestions:
1. Extract config loading into a separate function
2. Add error context wrapping
3. Consider using sync.Once for singleton instances
...

Example 2: Database Queries with SQLite Server

Query a database during conversation:

cat > mcp.json << 'EOF'
{
  "ollama": {
    "model": "qwen2.5:7b",
    "systemPrompt": "You are a skilled data analyst. Query databases to find insights, trends, and answers. Use SQLite tools to retrieve and analyze data. Present findings clearly with context and business implications."
  },
  "mcpServers": {
    "sqlite": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-sqlite", "./app.db"]
    }
  }
}
EOF

mcp-setu chat

In the chat:

❯ how many users registered in the last 7 days?

⚙ sqlite › query {"sql": "SELECT COUNT(*) FROM users WHERE created_at > datetime('now', '-7 days')"}
↳ query [{"COUNT(*)": 42}]

setu: There were 42 new user registrations in the last 7 days. This is up from 38 last week.

Example 3: Context Storage with Memory Server

Maintain context across conversations:

cat > mcp.json << 'EOF'
{
  "ollama": {
    "model": "llama3.2:3b",
    "systemPrompt": "You are a helpful assistant with persistent memory across sessions. Use the memory tool to save important context and retrieve previous information. Maintain continuity and remember user preferences and project details."
  },
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-memory"]
    }
  }
}
EOF

mcp-setu chat

Session 1:

❯ remember my project is a Go CLI tool

⚙ memory › set_context {"key": "project", "value": "Go CLI tool"}
↳ set_context [success]

❯ exit

Session 2 (same config):

❯ what was my project about?

⚙ memory › get_context {}
↳ get_context [{"key": "project", "value": "Go CLI tool"}]

setu: Your project is a Go CLI tool.

Example 4: Using Claude Desktop Config

Reuse your existing Claude Desktop setup:

# macOS
mcp-setu --config ~/Library/Application\ Support/Claude/claude_desktop_config.json chat

# Or add to your shell profile for easy access
alias claude-setu='mcp-setu --config ~/Library/Application\ Support/Claude/claude_desktop_config.json'

Example 5: Verbose Mode for Debugging

See tool calls and LLM iterations inline in the chat TUI:

mcp-setu chat --verbose

Output panel (above the input line):

you   list files in current directory
  💭 Processing...
  ⚙ read_directory
  ↳ {"contents":[{"name":"main.go","size":2048},…]}
setu  Here are the files in the current directory:
      - main.go (2 KB)
      - config.json (512 B)
      - README.md (4 KB)

Without --verbose, the 💭/⚙/↳ lines are hidden and you see only the final assistant message.

Example 6: Model Switching

Switch between models during chat:

mcp-setu chat

In chat:

❯ /model

Current model: llama3.2:3b
Available models:
  ✓ llama3.2:3b (tool support)
  ✓ qwen2.5:7b (tool support)
  - llama2:7b (no tool support)
  ✓ mistral-nemo:12b (tool support)

❯ /model qwen2.5:7b

✓ Switched to model: qwen2.5:7b

Example 7: Multi-Server Setup

Use multiple MCP servers together:

{
  "ollama": {
    "model": "qwen2.5:7b"
  },
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "."]
    },
    "sqlite": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-sqlite", "./data.db"]
    },
    "memory": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-memory"]
    }
  }
}

In chat:

❯ read config.json and remember its database path

⚙ filesystem › read_file {"path": "config.json"}
↳ read_file {"database": "./data.db"}

⚙ memory › set_context {"key": "db_path", "value": "./data.db"}
↳ set_context [success]

setu: I've read your config and saved the database path to memory.

❯ how many records are in the main table?

⚙ memory › get_context {}
↳ get_context [{"key": "db_path", "value": "./data.db"}]

⚙ sqlite › query {"sql": "SELECT COUNT(*) as count FROM main"}
↳ query [{"count": 1523}]

setu: There are 1,523 records in the main table.

Example 8: Performance Monitoring

Check performance metrics:

mcp-setu chat

In chat (after several exchanges):

❯ /stats

Performance Statistics
  Messages: 12
  Tool calls: 8
  Iterations: 5
  Session duration: 2m 34s
  Average response: 1.2s
  Longest response: 3.4s

Tips & Tricks

1. Override model temporarily without changing config:

   mcp-setu chat --model llama3.3:70b

2. Override system prompt:

   mcp-setu chat --system "You are a Python expert"

3. Validate before running:

   mcp-setu validate

4. List all tools available:

   mcp-setu tools

5. For non-interactive use, query tool metadata or run validation:

   mcp-setu tools          # list every tool
   mcp-setu validate       # exit non-zero on config/connection errors

   The chat command itself requires an interactive terminal (TUI).

Next Steps

• Configuration — More config options
• Troubleshooting — Solve issues
• CLI Reference — Full command documentation