Refusal Desk — pre-execution task-brief tribunal
Server Details
The Refusal Desk builds the no — a pre-execution tribunal that screens an incoming task brief
- Status
- Healthy
- Uptime
- 100.0% over the last 24 hours
- Last Tested
- Transport
- Streamable HTTP · MCP 2024-11-05
- URL
- ChatGPT
- –
- Claude
- –
TDQS
Scored across 4 tools
Each tool maps to a distinct action in the tribunal workflow: screen a brief, search duplicates, settle a verdict, or report calibration. check_brief and duplicate_lookup both accept a brief, but their purposes (screening vs. semantic duplicate detection) are clearly separated.
All names use snake_case, which is consistent. However, calibration_report and duplicate_lookup are noun phrases while check_brief and settle_outcome follow verb_noun, so the convention is mostly rather than fully uniform.
Four tools is well-scoped for a focused pre-execution tribunal. Each tool covers a necessary step without redundancy or bloat.
The surface covers the core lifecycle: check brief, look up duplicates, settle outcome, and report calibration. Minor gaps exist around retrieving or listing historical verdicts beyond the calibration report, but core workflows are supported.
Available Tools
4 toolscalibration_reportCInspect
Own per-class calibration + miss accounting
Inputs: no named parameters.
check_briefCInspect
Screen a task brief before execution {brief, standing_rules?} — $0.01 / trial grant / Fleet quota (same billing as HTTP)
Inputs: brief (required), standing_rules.
duplicate_lookupCInspect
Cross-requester semantic duplicate search {brief} — +$0.02 x402
Inputs: brief (required).
settle_outcomeCInspect
Settle a verdict {verdict_id, outcome, failure_class?, result?, cost_usd?} — idempotent; a forecast miss prices $0.05 against the per-principal per-period cap
Inputs: outcome (required), verdict_id (required), result, cost_usd, +1 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Related MCP Connectors
The decision layer for AI agents: act, escalate or refuse, and every decision comes back signed.
Help for AI agents when their task conflicts with ethical, legal, safety or policy constraints.
Human approval for AI agents: one ask, one person, one boolean.
Preventive human-approval write-gate for AI agents: writes commit only after a human approves.
Related MCP Servers
- AlicenseAqualityBmaintenanceExposes a decision tool that an agent calls before running a tool, sending the pending step to a decision model that assesses risk and completion and returns an allow or a reasoned refusal. It also works offline with a rules baseline that blocks catastrophic operations such as recursive deletes, downloads piped into a shell, and table drops.1MIT
- AlicenseNot gradedqualityBmaintenanceGives AI coding agents a local, mechanical workflow authority: an agent's issue claim is recorded, every subsequent write and irreversible step (push, merge, tag, release) is re-verified against that record, and review verdicts are bound to exact commit bytes so a later push invalidates them. Refusals report what was read and what to do next, keeping an unreadable tracker distinct from a forbidden action.1MIT
- AlicenseNot gradedqualityBmaintenanceThe kernel that doesn't believe the agents — a domain-free trust substrate for fleets of autonomous agents: verify what shipped, arbitrate collisions, refuse with structured reasons.116 PyPI22MIT
- AlicenseNot gradedqualityCmaintenanceGoverned robot task authority — signed mandates, bounded envelopes, evidence for every task.MIT
Glama MCP Gateway
Add one secure layer between your agents and this server.