Result Handler

Shows how Delta-MCP’s result handler automatically protects the LLM context window from large tool outputs.

What it demonstrates

Scenario	Raw output	Handled output
Large string	50 KB text	{ truncated: true, preview, totalChars, estimatedTokens }
Long array	50 items	{ paginated: true, items: [first 5], page, totalPages, hasMore }
Big object	100 keys	{ _summarized: true, _totalKeys: 100, key_0: "value_0", ... }
Upstream 429	thrown error	{ type: "rate_limited", retryAfterSeconds, upstream }

Run

npx tsx examples/result-handler/index.ts

Configuration

Set limits in the server constructor — they apply to every tool call:

new MyServer({
  resultHandler: {
    maxTokens: 2000,    // truncate strings over ~2000 tokens (~8 KB)
    paginateAfter: 50,  // paginate arrays longer than 50 items
  },
});

The model controls pagination by passing page and pageSize in tool args:

// model calls this after seeing hasMore: true in the previous result
await client.callTool("list_records", { page: 2, pageSize: 5 });

Rate-limit results give the model enough information to retry:

{
  "type": "rate_limited",
  "retryAfterSeconds": 30,
  "upstream": "call_rate_limited_api"
}

View on GitHub →

See also

• Result Handler — how it works
• Pagination example for a more involved pagination walkthrough