agenite

Agenite provides a modular, type-safe framework for building AI agents with TypeScript, including an MCP (Model Context Protocol) client package that allows connecting to standardized data sources and tools. This MCP setup enables you to interact with LLMs, invoke defined tools, validate inputs with JSON Schema, and manage tool results within a bidirectional, step-based execution model. The framework emphasizes strong type safety, clear tool definitions, and provider-agnostic integration, so you can plug in OpenAI, Anthropic, AWS Bedrock, or Ollama providers while coordinating complex tool usage through the MCP interface. As an MCP user, you can connect your agent-enabled server to data sources, web content, filesystems, and databases, enabling LLMs to access and act on real-world data via well-defined tool calls and schemas. To get started, install the MCP client package and configure your server to expose MCP endpoints that clients can connect to for data access and tool execution. Once connected, you can craft agents and tools that leverage strict input validation, reliable error handling, and a consistent API across providers. The tooling ecosystem in Agenite, including the @agenite/tool and @agenite/llm packages, lets you define tools with schemas and handlers, wire them into agents, and run multi-step workflows with streaming support if needed.

Prerequisites:

• Node.js (14.x or newer) and npm or pnpm
• Basic TypeScript tooling and a modern JS/TS build setup

Install the MCP client package (and any required dependencies):

# Using npm
npm install @agenite/mcp

# Or usingpnpm
pnpm add @agenite/mcp

If you plan to run an Agenite MCP server alongside, install core packages and a provider of choice (examples shown below).

# Core packages for agent orchestration and tooling
npm install @agenite/agent @agenite/tool @agenite/llm

# Provider example (choose at least one)
npm install @agenite/openai
# or
npm install @agenite/bedrock

Configure your project to build and run the MCP server. Typical steps:

1. Initialize a project (if starting from scratch)

mkdir agenite-mcp-server
cd agenite-mcp-server
npm init -y

2. Add TypeScript configuration and build scripts as needed for your server entry point.
3. Create your MCP server entry file (e.g., server.ts or server.js) and wire the MCP client to connect a client and expose endpoints according to your MCP server framework.
4. Build and run the server:

# Build (if using TypeScript)
npx tsc

# Run
node dist/server.js

Note: The exact server bootstrap code will depend on your MCP server implementation and how you expose the MCP protocol endpoints to clients. Ensure environment variables for provider access (e.g., OpenAI keys) are set as needed.

Tips and common considerations:

• Ensure your MCP client version aligns with the server's MCP protocol version to avoid compatibility issues.
• Use JSON Schema validation for tool inputs to improve reliability and error handling.
• When running locally, consider enabling streaming for real-time feedback from LLM calls and tool executions.
• If integrating multiple providers, centralize credentials and use provider-agnostic interfaces to simplify switching between providers.
• For debugging, enable verbose logging and leverage Agenite's pretty-logger middleware to get structured logs of LLM calls and tool results.