Quick Start

AI Client
Quick Start

Connect your AI client to OpenAgent Core in two ways — standard direct tool calls or the recommended code execution approach.

Standard Connection
Code Execution
Recommended
Option 1

Standard Connection

Direct tool calls via the standard MCP protocol. Point your client at the endpoint and start calling tools immediately.

Endpoint

https://your-mcp-server.workers.dev/mcp

Authentication

Authorization: Bearer <TENANT_API_KEY>
Example

Claude Desktop Configuration

Add the following to your Claude Desktop claude_desktop_config.json:

claude_desktop_config.json
{
  "mcpServers": {
    "openagent-core": {
      "url": "https://your-mcp-server.workers.dev/mcp",
      "headers": {
        "Authorization": "Bearer sk_live_your_tenant_api_key"
      }
    }
  }
}

The client loads all 34+ tool definitions via tools/list, then invokes them with tools/call.

Limitations of this approach

Context Overflow

All tool definitions load into the agent's context upfront, consuming hundreds of thousands of tokens before any work begins.

Expensive Data Flow

Large datasets — transactions, account histories — pass through the model multiple times, increasing cost and latency.

Poor Scaling

As more tools and data are used, token usage and response times grow quickly.

Recommended

Code Execution

The production-ready approach for large-scale financial data workflows with the MCP server.

How It Works

What Is Code Execution?

Instead of loading all tools as direct tool calls, you present the MCP server as a code API. The agent writes code (e.g., TypeScript) that calls the OpenAgent Core tools, loads only what it needs, and processes data in the execution environment before returning results.

Tools are discovered on-demand by exploring a file tree, and large datasets are filtered or aggregated in code — never passed wholesale through the model.

Benefits

Key Advantages

~98% token reduction

Load only the tools you need, not all 34+

Lower latency

Fewer model round-trips, faster responses

Handles large financial datasets

Transactions filtered and aggregated in code

On-demand tool discovery

Explore a file tree, read only what's relevant

Production-ready scaling

Token usage stays flat as tools and data grow

Ready to get started? The Code Execution Integration Guide covers setup, file structure, examples, and best practices in full detail.

Summary

Choose Your Approach

Compare the two connection methods side by side.

ApproachBest ForToken Efficiency
Standard (direct tools)
Quick tests, small workflows
Lower
Code execution
Recommended
Production, large datasets, 34+ tools
Much higher

Ready to build with the recommended approach?