← Agent workflows

Context-efficient tool routing

Tool catalog context budget auditor

Measure exact canonical UTF-8 bytes, map declared capability coverage, and produce a deterministic greedy subset without claiming semantic equivalence or global optimality.

Explicit-submit workbench

Run submitted JSON locally

Parsing and transformation happen in this browser. Nothing runs when a sample loads, and no input is fetched, uploaded, stored, or retained.

Load the inert sample or enter JSON, then explicitly run.

Hosted API and MCP parity

The same tool ID is available through POST /api/run and the fixed MCP run_tool method. Hosted requests are response-only. This browser workbench does not call either endpoint.

{
  "tool_id": "tool-catalog-context-budget-auditor",
  "input": "<SCHEMA_VALID_INPUT>",
  "response_mode": "full"
}