Rosbag MCP

警告 此代码由ai创建

ROSBag MCP服务器

该代码的灵感来自 ROSBag MCP服务器：使用LLM分析机器人数据，用于代理栓塞AI应用.

MCP（模型上下文协议）服务器，用于使用LLM分析ROS和ROS 2包文件。实现与机器人数据集的自然语言交互，用于调试、性能评估和提取见解。

特性

核心数据访问和管理

消息检索

数据过滤和导出

领域特定分析

导航分析

统计和事件检测

可视化和绘图

支持格式

安装

来自PyPI

# Using uv (recommended)
uv pip install rosbag-mcp

# Using pip
pip install rosbag-mcp

来自GitHub

# Using uv
uv pip install "rosbag-mcp @ git+https://github.com/cjh1995-ros/rosbag-mcp.git"

# Using pip
pip install "rosbag-mcp @ git+https://github.com/cjh1995-ros/rosbag-mcp.git"

来源（开发）

git clone https://github.com/cjh1995-ros/rosbag-mcp.git
cd rosbag-mcp
uv pip install -e ".[dev]"

需求

• Python>=3.10
• 主控程序 >= 1.0.0
• 玫瑰花包 >= 0.9.0
• numpy、matplotlib、枕头

用法

克劳德桌面/克劳德代码

增添 ~/Library/Application Support/Claude/claude_desktop_config.json （macOS）或 %APPDATA%\Claude\claude_desktop_config.json （Windows）：

{
  "mcpServers": {
    "rosbag": {
      "command": "rosbag-mcp"
    }
  }
}

或跑步 uvx 不安装：

{
  "mcpServers": {
    "rosbag": {
      "command": "uvx",
      "args": ["rosbag-mcp"]
    }
  }
}

作为MCP服务器（stdio）

rosbag-mcp

程序化使用（Python）

import asyncio
from mcp import ClientSession, StdioServerParameters
from mcp.client.stdio import stdio_client

async def analyze_bag():
    server_params = StdioServerParameters(
        command="python",
        args=["-m", "rosbag_mcp.server"],
    )
    
    async with stdio_client(server_params) as (read, write):
        async with ClientSession(read, write) as session:
            await session.initialize()
            
            # Set bag path
            await session.call_tool("set_bag_path", {
                "path": "/path/to/your/bag.bag"
            })
            
            # Get bag info
            result = await session.call_tool("bag_info", {})
            print(result.content[0].text)
            
            # Analyze trajectory
            result = await session.call_tool("analyze_trajectory", {
                "pose_topic": "/odom"
            })
            print(result.content[0].text)

asyncio.run(analyze_bag())

查询示例

一旦连接到Claude或另一个支持MCP的LLM：

"What bags do you have in /path/to/rosbags?"

"Show me the bag info for test.bag"

"What's the message structure of /odom topic?"

"Plot the robot's trajectory from the /odom topic"

"What was the robot's maximum velocity between t=10s and t=30s?"

"Has the robot ever passed close to position (x=2, y=-2) within 0.5 meters?"

"Analyze the LiDAR scan at t=15s - are there any obstacles within 2 meters?"

"Plot the commanded vs actual velocities for the first 30 seconds"

"Show me the TF tree - what are all the coordinate frames?"

"Filter the bag to only include /odom and /cmd_vel topics"

"Analyze the IMU data - what's the average acceleration?"

"Did the robot ever hit the costmap during navigation?"

"How well did the robot follow the planned path? What was the cross-track error?"

"Compare the commanded velocity vs actual velocity - what's the correlation?"

"When did the robot stop moving? Detect all stoppages."

"What's the publishing frequency of /odom? Are there any gaps?"

"Export the odometry data to CSV for analysis in Python"

"Is the robot experiencing wheel slip? Compare commanded vs actual velocity."

"Give me a navigation health report - any errors, recoveries, or planning failures?"

"Track the LiDAR minimum distance over time - when was the closest approach?"

"Plot a comparison of commanded velocity vs actual velocity with the difference."

"Detect all events where velocity dropped below -0.05 m/s (backward motion)."

输出示例

行李信息

{
  "path": "/data/robot_nav.bag",
  "duration": 73.97,
  "start_time": 1769892473.39,
  "end_time": 1769892547.36,
  "message_count": 191336,
  "topics": [
    {"name": "/odom", "type": "nav_msgs/msg/Odometry", "count": 4429},
    {"name": "/scan", "type": "sensor_msgs/msg/LaserScan", "count": 1108},
    {"name": "/tf", "type": "tf2_msgs/msg/TFMessage", "count": 27955}
  ]
}

轨迹分析

{
  "total_distance_m": 38.26,
  "duration_s": 73.87,
  "x_range": {"min": -19.75, "max": 5.21},
  "y_range": {"min": -4.20, "max": 6.53},
  "linear_speed": {"mean": 0.518, "max": 0.905, "min": 0.0},
  "angular_speed": {"mean": 0.178, "max": 1.058, "min": 0.0}
}

激光雷达分析

{
  "timestamp": 1769892483.52,
  "total_rays": 689,
  "valid_rays": 426,
  "statistics": {
    "min_distance": 4.12,
    "max_distance": 21.98,
    "mean_distance": 9.89
  },
  "obstacles": {
    "threshold_m": 2.0,
    "count": 0,
    "closest_distance": null
  }
}

IMU分析

{
  "topic": "/base_board/imu",
  "message_count": 7386,
  "duration_s": 73.88,
  "sample_rate_hz": 99.97,
  "linear_acceleration": {
    "x": {"mean": -0.0015, "std": 0.0205, "min": -0.2256, "max": 0.2467},
    "y": {"mean": -0.0018, "std": 0.0409, "min": -0.3311, "max": 0.4044},
    "z": {"mean": -0.0307, "std": 0.0916, "min": -0.9011, "max": 0.7922},
    "magnitude": {"mean": 0.0657, "max": 0.9559}
  },
  "angular_velocity": {
    "x": {"mean": 0.002, "std": 0.0459, "max_abs": 0.5867},
    "y": {"mean": 0.0001, "std": 0.0204, "max_abs": 0.2222},
    "z": {"mean": 0.0828, "std": 0.2694, "max_abs": 1.061}
  }
}

路径跟踪分析

{
  "path_topic": "/move_base/GlobalPlanner/plan",
  "pose_topic": "/amcl_pose",
  "tracking_samples": 641,
  "cross_track_error": {
    "mean_m": 0.0161,
    "std_m": 0.0125,
    "min_m": 0.0,
    "max_m": 0.0817,
    "median_m": 0.0132,
    "p95_m": 0.0377
  },
  "path_completion": {
    "final_progress": 1.0,
    "max_progress": 1.0
  }
}

主题统计

{
  "topic": "/odom",
  "message_count": 4429,
  "duration_s": 73.873,
  "frequency": {
    "mean_hz": 59.94,
    "std_hz": 53116.3,
    "min_hz": 5.54,
    "max_hz": 419430.4
  },
  "interval": {
    "mean_ms": 16.683,
    "std_ms": 11.203,
    "min_ms": 0.002,
    "max_ms": 180.384
  },
  "gaps": {
    "count": 92,
    "threshold_ms": 50.049,
    "largest_gap_ms": 180.384
  }
}

成本图违规检查

{
  "costmap_topic": "/move_base/local_costmap/costmap",
  "pose_topic": "/amcl_pose",
  "cost_threshold": 253,
  "violations": {"count": 0, "samples": []},
  "cost_distribution": {"0-49": 626, "50-99": 15},
  "summary": "No costmap violations detected - Robot stayed in free space"
}

可视化示例

二维轨迹图

Trajectory

激光雷达扫描（极坐标图）

LiDAR

速度时间序列

Velocity

建筑

┌─────────────────┐     stdio      ┌──────────────────┐
│   MCP Client    │◄──────────────►│  ROSBag MCP      │
│  (Claude, etc)  │    JSON-RPC    │     Server       │
└─────────────────┘                └────────┬─────────┘
                                            │
                                            ▼
                                   ┌──────────────────┐
                                   │  rosbags library │
                                   │  (Python)        │
                                   └────────┬─────────┘
                                            │
                    ┌───────────────────────┼───────────────────────┐
                    ▼                       ▼                       ▼
              ┌──────────┐           ┌──────────┐           ┌──────────┐
              │  .bag    │           │  .mcap   │           │  .db3    │
              │ (ROS 1)  │           │ (ROS 2)  │           │ (ROS 2)  │
              └──────────┘           └──────────┘           └──────────┘

工具架构示例

{
  "name": "get_messages_in_range",
  "description": "Get all messages from a topic within a time range",
  "inputSchema": {
    "type": "object",
    "properties": {
      "topic": {
        "type": "string",
        "description": "ROS topic name"
      },
      "start_time": {
        "type": "number",
        "description": "Start unix timestamp in seconds"
      },
      "end_time": {
        "type": "number",
        "description": "End unix timestamp in seconds"
      },
      "max_messages": {
        "type": "integer",
        "description": "Maximum messages to return (default: 100)"
      },
      "bag_path": {
        "type": "string",
        "description": "Optional: specific bag file"
      }
    },
    "required": ["topic", "start_time", "end_time"]
  }
}

参考文献

• 模型上下文协议（MCP）
• rosbags-纯Python ROS包库
• ROSBag MCP纸

许可证

MIT许可证