Loading ToolBench

Reading the current corpus…

rentahuman-mcp MCP server · C · ToolBench

rentahuman-mcp

MCP server for RentAHuman (workman.rent) — let an AI agent hire an identity-verified person for a short real-world task. Non-custodial payment, no escrow.

2025-06-18+STDIO5 tools

64/100

Definition Quality86

Protocol Readiness45

Supportability40

Spec posture

Inferred effective spec 2025-06-18+·current baseline 2026-07-28

Static source inference · medium confidence · evidence: structured output

No deprecated protocol patterns detected

Summary

Strong foundation with well-structured schemas, clear descriptions, and thoughtful safety constraints. All 5 tools have explicit Zod schemas with type definitions and parameter descriptions. Tool names follow verb_noun convention (search_tasks, create_task, etc.). Descriptions are detailed and explain WHAT, WHEN, and WHY. However, output schemas are documented only in descriptions (not as formal JSON Schema), and error handling lacks structured recovery guidance. Tool annotations present (readOnlyHint, destructiveHint, idempotentHint, openWorldHint) demonstrate spec awareness. Rate limiting and audit logging show production maturity. Main gaps: no formal output schema definitions, limited error recovery hints, and STDIO transport caps protocol readiness.

Tools (5)

assign_taskwriteauthsource verified82/100

MOCK: simulates an executor accepting the task. Dispatch to a real person is gated; until that gate opens a synthetic executor is used. Moves the task open -> assigned. Returns: { assignment_id, task_id, status, synthetic_executor }

confirm_taskwriteauthsource verified83/100

The agent confirms completion, which produces a SIMULATED non-custodial split instruction (executor + platform fee). There is NO real transaction yet (testnet/gate). The task moves to done. Returns: { task_id, status, payment_id, split: { network, amount_usdc, executor_usdc, fee_usdc, simulated:true } }

create_taskwriteauthsource verified88/100

Create a task for a human to perform. A stated PURPOSE is REQUIRED — anonymous or unmotivated tasks are rejected. Moderation is MANDATORY (the purpose is moderated too): banned or unsafe tasks are refused with 422 and never published. The precise point stays private until the task is accepted. Returns: { id, status, coarse_h3, moderation } Error: "Task rejected by moderation: ..." when the task is refused.

get_task_statusread onlyauthsource verified87/100

Current task status (coarse zone only, never precise coordinates). Read-only. Returns: { id, title, status, coarse_h3, budget_usdc }

search_tasksread onlyauthsource verified90/100

Find open tasks near a point. Returns ONLY a coarse zone (coarse_h3, about 5 km) and a distance — precise coordinates are never exposed before a task is accepted (privacy). Read-only. Returns: { count, tasks: [{ id, title, description, coarse_h3, budget_usdc, status, kind, distance_m }] }

Issues (5)

HIGHresponse-shaper ↗

Output schemas documented only in descriptions, not as formal JSON Schema. LLMs cannot parse unstructured text to infer response structure for downstream tool chaining.

search_taskscreate_taskassign_taskconfirm_taskget_task_status

HIGHrecovery-guide ↗

Error handling returns generic fail() messages without structured recovery guidance. 'Task rejected by moderation' lacks actionable next steps (e.g., 'Revise purpose field and retry').

create_task

CRITICALtransport-selection ↗

STDIO-only transport. Not remotely accessible; cannot be used by hosted MCP clients or cloud-based agents. Limits deployment to local/containerized environments.

all

MEDIUMtool-description ↗

assign_task and confirm_task descriptions note 'MOCK' and 'SIMULATED' but lack clarity on when real execution gates open. Agents may misunderstand production readiness.

assign_task

Recommendations

Define formal JSON Schema output types for all tools. Example: search_tasks should return { type: 'object', properties: { count: { type: 'integer' }, tasks: { type: 'array', items: { type: 'object', properties: { id, title, description, coarse_h3, budget_usdc, status, kind, distance_m } } } } }. This enables LLM reasoning about response structure.
Enhance error messages with structured recovery guidance. Instead of 'Task rejected by moderation: ...', return 'Task rejected by moderation: [reason]. Suggested fix: [action]. Retry with revised [field].' Example: 'Purpose too vague. Suggested fix: Explain the real-world use case (e.g., "Verify business hours for compliance audit"). Retry with a purpose 10+ characters.'
Migrate from STDIO to Streamable HTTP transport. STDIO limits deployment to local environments and breaks compatibility with hosted MCP clients. HTTP enables cloud-native deployment and multi-client access.
Clarify mock vs. production gates in tool descriptions. Add explicit statement: 'This tool is currently MOCKED (synthetic executor, no real dispatch). Real human dispatch is gated behind [condition]. Check [status endpoint] to verify gate status.' This prevents agent confusion about execution semantics.
Add batch variants: create_tasks (array of task objects) and confirm_tasks (array of task_ids). Batch operations reduce token overhead and latency for multi-task workflows common in agent loops.

Score history

Overall score trend

First recorded score · v2 rubric

64/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	C	64	2025-06-18+	v2