agentic -client

Agentic MCP Client is a standalone agent runner designed to execute tasks by leveraging MCP tools through multiple model providers (including Anthropic Claude, AWS Bedrock, and OpenAI). The client orchestrates a task by initializing MCP-enabled tools, sending a task to a chosen language model, and processing the model’s responses. If the model decides to call a tool, the client executes the corresponding tool (e.g., a Docker-based MCP service) and feeds the results back to the model, continuing until the task completes or a maximum iteration limit is reached. This enables autonomous agents to operate across cloud environments and interact with systems securely.

Prerequisites:

• Docker installed (for MCP tool containers) or a Python environment capable of running the client
• Access to a machine with network access to your desired MCP tool containers and LLM provider

Installation steps:

1. Clone the repository:

git clone https://github.com/peakmojo-agentic-mcp-client.git
cd peakmojo-agentic-mcp-client

2. Install Python dependencies (if using a Python environment):

pip install -r requirements.txt

3. Install/prepare the MCP client runtime (as described in the README):

• If using uv (Python):

uv sync

4. Prepare configuration files:

• Create a config.json in the project root defining your inference server settings and available MCP tools (as shown in the README example).

5. Run the agent worker:

uv run agentic_mcp_client/agent_worker/run.py

6. Start the dashboard (optional):

cd dashboard
npm i
npm run dev

Prerequisites recap:

• Python anduvx environment (for uv-based commands) or Docker for MCP tools
• Access to LLM providers (Anthropic/OpenAI Bedrock) as configured in config.json
• Docker if using containerized MCP tools

Tips and common considerations:

• The MCP tool definitions in config.json are typically Docker-based containers. Ensure the image names (e.g., buryhuang/mcp-remote-macos-use:latest) exist and are accessible from your environment.
• When using Bedrock or other LLM providers, securely store API keys (e.g., in config.json or environment variables) and avoid committing secrets to version control.
• The agent supports a configurable max_iterations to control how long the task runs; adjust this to balance thoroughness with resource usage.
• If you encounter connection issues to LLM providers, verify network access, API keys, and base_url endpoints in inference_server.
• The dashboard provides a UI for monitoring tasks and agent progress; ensure Node.js and dependencies are installed if you plan to use it.