ContextQMD
Back to Openclaw

Tool Search

docs/tools/tool-search.md

2026.5.127.8 KB
Original Source

Tool Search is an experimental OpenClaw PI-agent feature. It gives PI agents one compact way to discover and call large tool catalogs. It is useful when the run has many available tools but the model is likely to need only a few of them.

This page documents OpenClaw PI Tool Search. It is not the Codex-native tool search or dynamic-tools surface. Codex-native code mode, tool search, deferred dynamic tools, and nested tool calls are stable Codex harness surfaces and do not depend on tools.toolSearch.

When enabled for PI, the model receives one tool_search_code tool by default. That tool runs a short JavaScript body in an isolated Node subprocess with an openclaw.tools bridge:

js
const hits = await openclaw.tools.search("create a GitHub issue");
const tool = await openclaw.tools.describe(hits[0].id);
return await openclaw.tools.call(tool.id, {
  title: "Crash on startup",
  body: "Steps to reproduce...",
});

The catalog can include OpenClaw tools, plugin tools, MCP tools, and client-provided tools. The model does not see every full schema up front. Instead, it searches compact descriptors, describes one selected tool when it needs the exact schema, and calls that tool through OpenClaw.

Codex harness runs do not receive these experimental OpenClaw Tool Search controls. OpenClaw passes product capabilities to Codex as dynamic tools, and Codex owns the stable native code mode, native tool search, deferred dynamic tools, and nested tool calls.

How a turn runs

At planning time the PI embedded runner builds the effective catalog for the run:

  1. Resolve the active tool policy for the agent, profile, sandbox, and session.
  2. List eligible OpenClaw and plugin tools.
  3. List eligible MCP tools through the session MCP runtime.
  4. Add eligible client tools supplied for the current run.
  5. Index compact descriptors for search.
  6. Expose either the PI code bridge or the structured fallback tools to the model.

At execution time every real tool call returns to OpenClaw. The isolated Node runtime does not hold plugin implementations, MCP client objects, or secrets. openclaw.tools.call(...) crosses the bridge back into the Gateway, where the normal policy, approval, hook, logging, and result handling still apply.

Modes

tools.toolSearch has two model-facing modes:

  • code: exposes tool_search_code, the default compact JavaScript bridge.
  • tools: exposes tool_search, tool_describe, and tool_call as plain structured tools for providers that should not receive code.

Both modes use the same catalog and execution path. The only difference is the shape the model sees. If the current runtime cannot launch the isolated Node code-mode child process, the default code mode falls back to tools before catalog compaction.

Both modes are experimental. Prefer direct tool exposure for small PI tool catalogs, and prefer the Codex-native stable surfaces for Codex harness runs.

There is no separate source-selection config. When Tool Search is enabled, the catalog includes eligible OpenClaw, MCP, and client tools after normal policy filtering.

Why this exists

Large catalogs are useful but expensive. Sending every tool schema to the model makes the request larger, slows planning, and increases accidental tool selection.

Tool Search changes the shape:

  • direct tools: the model sees every selected schema before the first token
  • Tool Search code mode: the model sees one compact code tool and a short API contract
  • Tool Search tools mode: the model sees three compact structured fallback tools
  • during the turn: the model loads only the tool schemas it actually needs

Direct tool exposure is still the right default for small catalogs. Tool Search is best when one run can see many tools, especially from MCP servers or client-provided app tools.

API

openclaw.tools.search(query, options?)

Searches the effective catalog for the current run. Results are compact and safe to put back into prompt context.

js
const hits = await openclaw.tools.search("calendar event", { limit: 5 });

openclaw.tools.describe(id)

Loads full metadata for one search result, including the exact input schema.

js
const calendarCreate = await openclaw.tools.describe("mcp:calendar:create_event");

openclaw.tools.call(id, args)

Calls a selected tool through OpenClaw.

js
await openclaw.tools.call(calendarCreate.id, {
  summary: "Planning",
  start: "2026-05-09T14:00:00Z",
});

The structured fallback mode exposes the same operations as tools:

  • tool_search
  • tool_describe
  • tool_call

Runtime boundary

The code bridge runs in a short-lived Node subprocess. The subprocess starts with Node permission mode enabled, an empty environment, no filesystem or network grants, and no child-process or worker grants. OpenClaw enforces a parent-process wall-clock timeout and kills the subprocess on timeout, including after async continuations.

The runtime exposes only:

  • console.log, console.warn, and console.error
  • openclaw.tools.search
  • openclaw.tools.describe
  • openclaw.tools.call

Normal OpenClaw behavior still applies to final calls:

  • tool allow and deny policies
  • per-agent and per-sandbox tool restrictions
  • owner-only gating
  • approval hooks
  • plugin before_tool_call hooks
  • session identity, logs, and telemetry

Config

Enable Tool Search for PI runs with the default code bridge:

bash
openclaw config set tools.toolSearch true

Equivalent JSON:

json5
{
  tools: {
    toolSearch: true,
  },
}

Use the structured fallback tools instead for PI runs:

json5
{
  tools: {
    toolSearch: {
      mode: "tools",
    },
  },
}

Tune code-mode timeout and search result limits:

json5
{
  tools: {
    toolSearch: {
      mode: "code",
      codeTimeoutMs: 10000,
      searchDefaultLimit: 8,
      maxSearchLimit: 20,
    },
  },
}

Disable it:

json5
{
  tools: {
    toolSearch: false,
  },
}

Prompt and telemetry

Tool Search records enough telemetry to compare it with direct tool exposure:

  • total serialized tool and prompt bytes sent to the harness
  • catalog size and source breakdown
  • search, describe, and call counts
  • final tool calls executed through OpenClaw
  • selected tool ids and sources

Session logs should make it possible to answer:

  • how many tool schemas the model saw up front
  • how many search and describe operations it performed
  • which final tool was called
  • whether the result came from OpenClaw, MCP, or a client tool

E2E validation

The gateway E2E runner proves both paths with the PI harness:

bash
node --import tsx scripts/tool-search-gateway-e2e.ts

It creates a temporary fake plugin with a large tool catalog, starts the mock OpenAI provider, starts a Gateway once in direct mode and once with Tool Search enabled, then compares provider request payloads and session logs.

The regression proves:

  1. Direct mode can call the fake plugin tool.
  2. Tool Search can call the same fake plugin tool.
  3. Direct mode exposes the fake plugin tool schemas directly to the provider.
  4. Tool Search exposes only the compact bridge.
  5. The Tool Search request payload is smaller for the large fake catalog.
  6. Session logs show the expected tool-call counts and bridged call telemetry.

Failure behavior

Tool Search should fail closed:

  • if a tool is not in the effective policy, search should not return it
  • if a selected tool becomes unavailable, tool_call should fail
  • if policy or approval blocks execution, the call result should report that block instead of bypassing it
  • if the code bridge cannot create an isolated runtime, use mode: "tools" or disable Tool Search for that deployment