Phospho

Provides a local MCP server to control phosphobot actions and stream camera frames.

python

8

GitHub Stars

python

Language

6 months ago

First Indexed

2 months ago

Catalog Refreshed

Documentation & install

Readme and setup notes from the catalogue, plus a client-ready config you can copy for your MCP host.

Installation

Add the following to your MCP client configuration file.

Configuration

View docs

{
  "mcpServers": {
    "phospho-app-phospho-mcp-server": {
      "command": "uv",
      "args": [
        "run",
        "mcp",
        "install",
        "server.py"
      ]
    }
  }
}

You deploy and run a dedicated MCP server that bridges your Claude-style client to phosphobot actions and camera streams. This server exposes tools to trigger robot actions and to fetch camera frames, enabling natural language control and real-time feedback for your robot setup.

How to use

You interact with the phospho MCP Server through your MCP client by calling its available tools. The two core capabilities you’ll use are:

get_camera_frame: capture a current image from phosphobot’s camera. Use this to display what the robot is seeing in your interface.
pickup_object: replay a pre-recorded episode to simulate grabbing a specified object. This is useful for testing object manipulation workflows.

In practice, you’ll describe the action you want in natural language to your MCP client, and the client will translate that into tool invocations such as obtaining a camera frame or launching a replay episode. Results from actions like the camera feed arrive as JPEG data that you can render in your UI, and replay results provide feedback on the simulated operation.

How to install

Prerequisites you should have before starting:

a working Node.js or Python environment (as required by the MCP server tooling in your setup)
the uv toolchain for running MCP servers (uvx)

Follow these steps to install and run the MCP server locally:

Install uv on macOS or Linux

# macOS
curl -LsSf https://astral.sh/uv/install.sh | sh

# Linux
curl -LsSf https://astral.sh/uv/install.sh | sudo bash

# Windows
powershell -ExecutionPolicy ByPass -Command "irm https://astral.sh/uv/install.ps1 | iex"

Clone the MCP server repository and install the server tooling

git clone https://github.com/phospho-app/phospho-mcp-server.git

cd phospho-mcp-server/phospho-mcp-server
uv run mcp install server.py

Start the MCP server in development mode to test and validate within your MCP client

uv run mcp dev server.py

Additional notes and endpoints

The MCP server communicates with phosphobot via its local REST API. Key endpoints you may encounter include the camera frame stream and the replay trigger: the camera frame fetch corresponds to acquiring a JPEG image from the robot’s camera, while the replay trigger initiates a dataset-driven action sequence.

System behavior you can expect: the server manages the phosphobot process and forwards tool calls to the local API, returning results back to your MCP client. If you change the port of the underlying phosphobot API, ensure you reflect that in your client configuration.

Tool references

The MCP server exposes the following tools you can invoke from your MCP client:

get_camera_frame — Captures a JPEG image from phosphobot’s camera. Returns a base64-encoded JPEG image.

pickup_object — Launches a replay episode to simulate object manipulation. Accepts object names such as banana or other predefined items.

Available tools

get_camera_frame

Captures a JPEG image from phosphobot's camera and returns a base64-encoded image.

pickup_object

Launches a replay episode to simulate object manipulation for a given object name.