zypher-agent

Zypher Agent exposes an MCP-enabled runtime that can orchestrate AI tasks within your application. It ships with a flexible tool system for file operations, search, and terminal commands, and supports integrating external MCP servers via OAuth authentication. Use the agent to run task-oriented sessions where the agent reason autonomously and issue tool-operations while you observe streaming events. The MCP integration lets the agent communicate with MCP servers to delegate reasoning steps, fetch model context, and persist checkpoints as the agent evolves tasks. To get started, configure the MCP server in your environment and supply the necessary OAuth credentials, then reference the MCP server in the agent configuration so it can route model-context interactions accordingly. As you work, you can leverage the built-in tools and post-inference interceptors to customize behavior, log progress, and enforce safety checks during task execution.

Prerequisites:

• Node.js (LTS) installed on your machine
• npm or yarn for package management
• Git for cloning repositories

1. Clone the repository or install the MCP server package (example assumes npm package availability): npm install -g @zypher/agent || yarn global add @zypher/agent

2. Install dependencies in your project (if integrating as a local module): git clone https://github.com/corespeed-io/zypher-agent.git cd zypher-agent npm install

3. Build the project (if applicable): npm run build

4. Run the MCP-enabled server (example): node dist/server.js

5. Set up environment variables for OAuth and model providers as shown in the mcp_config example. Ensure the MCP server is accessible from your application.

Prerequisites are Node.js and npm, plus credentials for OAuth and your chosen model provider.

Notes and tips:

• Ensure your OAuth credentials (OAUTH_CLIENT_ID and OAUTH_CLIENT_SECRET) are securely stored and not committed to version control.
• The MCP registry URL is used if you operate with a custom MCP server registry; default to the public MCP registry if applicable.
• Adjust LOG_LEVEL to debug during development to capture detailed traces, and switch to info or warning in production.
• The agent relies on a tool system (file operations, search, terminal commands). Extend or replace tools to fit your environment.
• If you encounter port conflicts, change the PORT environment variable accordingly.
• For troubleshooting, verify that the dist/server.js path exists after your build step; if your entry point differs, update the mcp_config args accordingly.
• When using multiple MCP servers, ensure proper OAuth scopes and role-based access for each server to avoid authorization failures.