Loading ToolBench
MCP server connector for Immersive Commons — bridges stdio MCP clients (Claude Desktop, Cursor, Windsurf) to the hosted Streamable HTTP server at https://www.immersivecommons.com/api/mcp (138 tools)
23 tools with mixed quality. Most have basic descriptions (10-50 chars, below the 50-200 char baseline for LLM optimization) and minimal parameter documentation. Input schemas present for ~13 tools but many lack type definitions or descriptions for parameters. No output schemas documented. Error handling absent. Tool names follow verb_noun convention (good), but descriptions are too terse to guide LLM selection. Example: 'List upcoming Immersive Commons events' (34 chars) lacks context on when to use it vs. other discovery tools. 'Queue an RSVP' (13 chars) is below the 20-char minimum and provides no guidance on prerequisites or side effects. No tool declares permissions, scopes, or idempotency guarantees.
Query the research RAG corpus
Run up to 20 public GET reads in one request
Queue a resource booking
Cancel the pending tier request
Read the public donor wall
Look up a single event by its Luma URL
Read the caller's own activity log
9 tools (batchPublicReads, requestEvent, listResources, getMyLeaderboardStatus, submitHighlightPending, revokeOwnToken, setupCheck, startSignup, getMyTier, requestTier, cancelTierRequest) have NO visible input schema or output schema documentation. Cannot verify parameter types or return structure.
Descriptions are too terse (avg 28 chars, baseline 50-200). Examples: 'Queue an RSVP' (13 chars), 'Toggle commit-leaderboard opt-in' (32 chars), 'Deterministic am I set up? probe' (32 chars). LLMs cannot infer when to select these tools vs. similar ones. No guidance on prerequisites, side effects, or idempotency.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
Read the caller's commit-leaderboard state
Read the caller's membership tier
List bookable resources
List upcoming Immersive Commons events
Poll a device-code grant for completion
Propose a member event (operator-approved)
Request a higher membership tier
Self-revoke the presenting token
Queue an RSVP to a Luma event
Search the member directory
Toggle commit-leaderboard opt-in
Deterministic 'am I set up?' probe
Begin the RFC 8628 device-code token mint
File feedback, a feature request, or a breakage report
Submit a highlight to the moderation queue
Cursor-tail the caller's agentic event log
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., event_id, resource_id). Breaks tool composition and forces discovery detours.
No error handling guidance. Tools like rsvpToEvent, bookResource, and requestEvent are write operations but provide no recovery hints (e.g., 'If email is invalid, call searchDirectory first'). No categorization of retryable vs. fatal errors.
No permission/scope declarations. Tools like revokeOwnToken and setLeaderboardOptIn are sensitive but do not declare required scopes (e.g., 'Requires: auth:revoke'). Agents cannot be configured with least-privilege permissions.