toolquest
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| TOOLQUEST_DISABLE_TRACES | No | Set to 1 to disable disk traces. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| list_roomsA | Discover all built-in rooms with their IDs, versions, difficulty, introductions, and par action counts. This does not create or change a run. |
| start_runA | Create an isolated deterministic escape-room run. Returns an opaque runId, initial stateVersion, room introduction, and next-step guidance. |
| get_runA | Resume a persisted run by reading its public current snapshot, status, stateVersion, stateHash, event count, and score when terminal. This does not append an event. |
| replay_runA | Deterministically rebuild a run from its redacted event log and verify every stateVersion, stateHash, outcome, and final state. This does not change the run. |
| export_reportA | Generate a Markdown benchmark report containing run metadata, score, replay verification, and a redacted event timeline. Returns content without writing a file. |
| lookA | Read the current room, visible object IDs, destination IDs, inventory, and attempts remaining. This does not change virtual world state. |
| inspectA | Inspect one visible object by targetId. Returns clues and valid interaction IDs without changing virtual world state. |
| moveA | Move to a destinationId returned by look. This changes virtual room state. Provide a unique actionId and the latest expectedStateVersion; exact retries with the same actionId are safe. |
| useA | Perform an interactionId returned by inspect, optionally with an inventory itemId. This changes only virtual room state and is protected by actionId and stateVersion. |
| submitA | Submit a candidate answer to the final room challenge. Incorrect answers may consume an attempt; a correct answer ends the run and returns a deterministic score. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 10 tools
Each tool has a clearly distinct role: run lifecycle tools (start/get/replay/export) are separate from in-room inspection tools (look/inspect) and state-changing actions (move/use/submit). There is no meaningful overlap between any pair of tools.
Most tools follow a clear verb_noun pattern such as list_rooms, start_run, get_run, and export_report. The in-room actions use single verbs like look, inspect, move, use, and submit, which is a minor deviation but still predictable and readable.
10 tools is well-scoped for an escape-room benchmark server. Each tool covers a necessary part of discovery, run lifecycle, in-room interaction, and reporting without redundancy.
The toolset covers the main workflow: discover rooms, start a run, observe state, act in rooms, submit answers, resume runs, replay, and export reports. Minor gaps like no run-listing or explicit run-abandon tool are not required for the core benchmark use case.