> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mcp-use.com/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP Logging

> Clean, configurable logging system designed specifically for MCP servers

<Card title="" img="https://cdn.mcp-use.com/docs/python/Logging.jpg" horizontal="true" />

<Info>
  mcp-use provides clean, informative logging about MCP operations, making it easy to understand server activity and debug issues in both production and development.
</Info>

MCP-specific logging provides detailed information about which MCP methods are being called, including the specific method name, session IDs, and execution details. MCP requests use a distinct `MCP:` prefix to differentiate them from regular HTTP logs.

## Why Use MCP-Specific Logging?

This type of logging is essential because:

* **Clear visibility** into which MCP methods are being called
* **Easy debugging** of MCP protocol interactions
* **Request tracking** with session IDs and method names
* **Production monitoring** of server usage patterns
* **Error tracking** for tools, resources, and prompts
* **Performance insights** with execution timing

## Log Levels

mcp-use provides three logging modes controlled by the `debug` parameter and `DEBUG` environment variable:

### Production Logs (Default: `debug=False`)

Clean, readable logs showing MCP method calls with the `MCP:` prefix:

```
MCP:  127.0.0.1:58478 - "POST /mcp [tools/call:search] HTTP/1.1" 200
MCP:  127.0.0.1:58478 - "POST /mcp [resources/list] HTTP/1.1" 200
MCP:  127.0.0.1:58478 - "POST /mcp [initialize] HTTP/1.1" 200
```

### Debug Logs (`debug=True`)

Same logs as production + development routes (inspector, docs). By default, inspector logs are hidden:

```
MCP:  127.0.0.1:58478 - "POST /mcp [tools/call:search] HTTP/1.1" 200
INFO: 127.0.0.1:58478 - "GET  /docs HTTP/1.1" 200
```

### Full Debug Logs (`DEBUG=2` environment variable)

Same logs as debug mode + JSON-RPC request/response logging:

```
MCP:  127.0.0.1:58478 - "POST /mcp [tools/call:search] HTTP/1.1" 200
MCP:  [tools/call:search] Request (45.2ms): {"jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": {"name": "search", "arguments": {"query": "test"}}}
MCP:  [tools/call:search] Response: {"jsonrpc": "2.0", "id": 1, "result": {"content": [...]}}
```

#### Pretty Print Mode (`pretty_print_jsonrpc=True`)

For easier debugging, enable pretty-printed JSON-RPC logs with Rich panels:

```python theme={null}
server = MCPServer("my-server", pretty_print_jsonrpc=True)
```

```
MCP:  127.0.0.1:58478 - "POST /mcp [tools/call:search] HTTP/1.1" 200
╭─ tools/call:search Request ─────────────────────── 45.2ms ──╮
│ {                                                           │
│   "jsonrpc": "2.0",                                         │
│   "id": 1,                                                  │
│   "method": "tools/call",                                   │
│   "params": {                                               │
│     "name": "search",                                       │
│     "arguments": {"query": "test"}                          │
│   }                                                         │
│ }                                                           │
╰─────────────────────────────────────────────────────────────╯
╭─ tools/call:search Response ────────────────────────────────╮
│ {                                                           │
│   "jsonrpc": "2.0",                                         │
│   "id": 1,                                                  │
│   "result": {"content": [...]}                              │
│ }                                                           │
╰─────────────────────────────────────────────────────────────╯
```

## Understanding the Log Format

Each MCP log line contains detailed information about the request:

```
MCP:  127.0.0.1:58478 - "POST /mcp [tools/call:search] HTTP/1.1" 200
```

Breaking down each component:

<Steps>
  <Step title="Log Prefix">
    <code style={{color: '#10b981'}}>MCP:</code> - Identifies this as an MCP-specific log (distinct from regular HTTP `INFO:` logs)
  </Step>

  <Step title="Client Address">
    <code style={{color: '#3b82f6'}}>127.0.0.1:58478</code> - The IP address and port of the client making the request
  </Step>

  <Step title="HTTP Method">
    <code style={{color: '#f59e0b'}}>POST</code> - The HTTP method used for the request
  </Step>

  <Step title="Endpoint">
    <code style={{color: '#8b5cf6'}}>/mcp</code> - The MCP server endpoint path
  </Step>

  <Step title="MCP Method">
    <code style={{color: '#06b6d4'}}>\[tools/call:search]</code> - The specific MCP method being called (e.g., tools/call, resources/list, etc.)
  </Step>

  <Step title="HTTP Version">
    <code style={{color: '#64748b'}}>HTTP/1.1</code> - The HTTP protocol version
  </Step>

  <Step title="Status Code">
    <code style={{color: '#10b981'}}>200</code> - The HTTP response status code
  </Step>
</Steps>

### Common MCP Methods

The logs clearly show which MCP methods are being called, such as:

* `[initialize]` - Server initialization
* `[tools/list]` - Listing available tools
* `[tools/call:search]` - Calling a specific tool
* `[resources/list]` - Listing available resources
* `[resources/read:config]` - Reading a specific resource
* `[prompts/get:assistant]` - Getting a specific prompt

## Configuration

Logging is automatically configured based on the `DEBUG` environment variable:

```bash theme={null}
# Production logging (default)
python server.py

# Debug mode with development routes (inspector, docs)
DEBUG=1 python server.py

# Full debug mode with JSON-RPC request/response logging
DEBUG=2 python server.py
```

Or programmatically:

```python theme={null}
# Production logging (default)
server = MCPServer("my-server")
server.run(transport="streamable-http")

# Debug mode with dev routes
server = MCPServer("my-server", debug=True)
server.run(transport="streamable-http")

# Debug mode with custom paths
server = MCPServer(
    name="my-server",
    debug=True,
    mcp_path="/api/mcp",  # Custom MCP endpoint
    docs_path="/custom-docs",
    inspector_path="/custom"
)
server.run(transport="streamable-http")

# Effective routes:
# - MCP:       /api/mcp
# - Docs:      /custom-docs
# - Inspector: /custom/inspector
```

## Logging Options

### Inspector Logs (`show_inspector_logs`)

By default, inspector-related logs (requests to `/inspector/*`) are hidden to keep the output clean. Enable them if needed:

```python theme={null}
# Hide inspector logs (default)
server = MCPServer("my-server", show_inspector_logs=False)

# Show inspector logs
server = MCPServer("my-server", show_inspector_logs=True)
```

**With `show_inspector_logs=False` (default):**

```
MCP:  127.0.0.1:58478 - "POST /mcp [initialize] HTTP/1.1" 200
MCP:  127.0.0.1:58478 - "POST /mcp [tools/list] HTTP/1.1" 200
```

**With `show_inspector_logs=True`:**

```
MCP:  127.0.0.1:58478 - "POST /mcp [initialize] HTTP/1.1" 200
INFO: 127.0.0.1:58478 - "GET /inspector HTTP/1.1" 200
INFO: 127.0.0.1:58478 - "GET /inspector/config.json HTTP/1.1" 200
MCP:  127.0.0.1:58478 - "POST /mcp [tools/list] HTTP/1.1" 200
```

### MCP Logs Only (`mcp_logs_only`)

By default, the server shows both `MCP:` protocol logs and `INFO:` HTTP access logs. Enable `mcp_logs_only` to suppress all HTTP access logs and only show MCP protocol traffic:

```python theme={null}
# Show all logs (default)
server = MCPServer("my-server", mcp_logs_only=False)

# Only show MCP protocol logs
server = MCPServer("my-server", mcp_logs_only=True)
```

**With `mcp_logs_only=False` (default):**

```
MCP:  127.0.0.1:58478 - "POST /mcp [initialize] HTTP/1.1" 200
INFO: 127.0.0.1:58478 - "POST /mcp HTTP/1.1" 200
MCP:  127.0.0.1:58478 - "POST /mcp [tools/list] HTTP/1.1" 200
INFO: 127.0.0.1:58478 - "POST /mcp HTTP/1.1" 200
INFO: 127.0.0.1:58478 - "GET  /docs HTTP/1.1" 200
```

**With `mcp_logs_only=True`:**

```
MCP:  127.0.0.1:58478 - "POST /mcp [initialize] HTTP/1.1" 200
MCP:  127.0.0.1:58478 - "POST /mcp [tools/list] HTTP/1.1" 200
```

This removes the duplicate `INFO:` lines for MCP requests and hides non-MCP HTTP traffic (`/docs`, `/inspector`, static files), giving you a clean view of only MCP protocol activity.

### Pretty Print JSON-RPC (`pretty_print_jsonrpc`)

Enable pretty printing to display JSON-RPC requests and responses as formatted Rich panels:

```python theme={null}
# Plain text JSON-RPC logs (default)
server = MCPServer("my-server", pretty_print_jsonrpc=False)

# Pretty-printed JSON-RPC logs with Rich panels
server = MCPServer("my-server", pretty_print_jsonrpc=True)
```

**With `pretty_print_jsonrpc=False` (default):**

```
MCP:  127.0.0.1:58478 - "POST /mcp [tools/call:search] HTTP/1.1" 200
MCP:  [tools/call:search] Request (45.2ms): {"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search"}}
MCP:  [tools/call:search] Response: {"jsonrpc":"2.0","id":1,"result":{...}}
```

**With `pretty_print_jsonrpc=True`:**

```
MCP:  127.0.0.1:58478 - "POST /mcp [tools/call:search] HTTP/1.1" 200
╭─ tools/call:search Request ─────────────────────── 45.2ms ──╮
│ {                                                           │
│   "jsonrpc": "2.0",                                         │
│   "id": 1,                                                  │
│   "method": "tools/call",                                   │
│   "params": {"name": "search"}                              │
│ }                                                           │
╰─────────────────────────────────────────────────────────────╯
```

### Combined Example

```python theme={null}
server = MCPServer(
    name="my-server",
    debug=True,
    show_inspector_logs=False,    # Hide inspector noise
    mcp_logs_only=True,           # Only show MCP protocol logs
    pretty_print_jsonrpc=True,    # Pretty-printed JSON-RPC panels
)
server.run(transport="streamable-http")
```

This makes it much easier to understand what's happening with your MCP server and debug any issues in both production and development environments.

## Client Log Messages (MCP Protocol)

In addition to server-side logging, MCP servers can send structured log messages to clients using the [MCP logging protocol](https://modelcontextprotocol.io/specification/2025-11-25/server/utilities/logging). This lets clients display or process server activity in their UI.

### Sending Log Messages

Use `ctx.log()` or the convenience methods inside tool handlers:

```python theme={null}
@mcp.tool()
async def process_data(ctx: Context, data: str) -> str:
    await ctx.debug("Starting data processing")
    await ctx.info("Processing 1000 records")
    await ctx.warning("Skipping 3 malformed records")

    try:
        result = do_work(data)
    except Exception as e:
        await ctx.error(f"Processing failed: {e}")
        raise

    return result
```

### Log Levels

Log levels follow [RFC 5424](https://datatracker.ietf.org/doc/html/rfc5424#section-6.2.1) syslog severity, from lowest to highest:

| Level       | Description             | Use Case            |
| ----------- | ----------------------- | ------------------- |
| `debug`     | Detailed debugging info | Function entry/exit |
| `info`      | Informational messages  | Progress updates    |
| `notice`    | Normal but significant  | Config changes      |
| `warning`   | Warning conditions      | Deprecated usage    |
| `error`     | Error conditions        | Operation failures  |
| `critical`  | Critical conditions     | Component failures  |
| `alert`     | Immediate action needed | Data corruption     |
| `emergency` | System unusable         | Complete failure    |

### Client-Controlled Filtering

Clients can set a minimum log level via `logging/setLevel`. When a client sets the level to `warning`, the server automatically suppresses `debug`, `info`, and `notice` messages  -  only `warning` and above are sent:

```
Client → Server:  logging/setLevel { level: "warning" }
Server → Client:  ✗ ctx.debug("...")     # suppressed
Server → Client:  ✗ ctx.info("...")      # suppressed
Server → Client:  ✓ ctx.warning("...")   # sent
Server → Client:  ✓ ctx.error("...")     # sent
```

This filtering happens automatically  -  you don't need to check the level yourself. Just call `ctx.log()` at the appropriate level and the server handles the rest.

<Tip>
  The default level is `debug` (all messages sent). Clients opt in to filtering by calling `logging/setLevel`.
</Tip>

## Next Steps

* [Inspector UI](/python/server/inspector) - Monitor logs in real-time
* [Running the Server](/python/server/running) - How to run your server
