Loading ToolBench
MCP server for AgentLair — exposes agent trust verification, PoPA streaks, and audit lookup as native tools for Claude desktop, Cursor, Cline, and any MCP client.
AgentLair MCP server exposes 5 well-named, domain-specific tools for agent trust verification and attestation lookup. All tools have clear, detailed descriptions (150 - 250 chars) explaining WHAT they do and WHEN to use them. Input schemas are present with proper type constraints (enums, regex patterns, min/max bounds). However, output schemas are not documented in the source, LLMs cannot see what fields to expect from responses. Error handling is minimal; no recovery guidance or actionable error messages visible. Tool names follow verb_noun convention (verify_, check_, get_, lookup_) and are unambiguous. Parameters are well-constrained (agent_id pattern, min_level enum, limit bounds). No security issues detected, no secrets in params, x402 payment gating is server-side. Composition is clean: each tool has one responsibility. Missing: output schema documentation, error recovery patterns, and per-tool risk annotations.
Fast-path trust check: does this agent meet a minimum ATF level? Use before granting access to a tool, capability, or paid action. Cheaper than the full profile — returns just the gate decision. Levels: intern (default minimum), junior, senior, principal. Same x402 pricing as verify_agent (0.01 USDC anonymous; free with AGENTLAIR_AAT).
Get the Proof of Persistent Activity streak for any DID. Returns: current streak in days, longest streak ever, total attestations, last attestation timestamp, gap count, genesis date, and the latest SCITT transparency-log entry. PoPA is the bond-of-presence — agents earn it by being active each day, signed and chained into a transparency log. Useful as a freshness signal: an agent attesting today is operationally alive today. Public endpoint. No auth required.
Top AgentLair agents by attestation count. Useful for discovering agents with the longest behavioral track records, or sanity-checking that the registry is alive. Public endpoint. No auth required.
Look up metadata for an AgentLair Agent Authentication Token (AAT) by its jti. Returns: issued_at, expires_at, audience, scopes, and live status (active | expired | revoked). Use this to verify a token presented by another agent — was it issued? is it still valid? has it been revoked? Tokens are auditable while valid and up to ~5 minutes after expiry. x402-gated at 0.001 USDC per lookup for anonymous callers. With AGENTLAIR_AAT set, the call is rejected — this endpoint is intentionally pay-as-you-go (payment IS authentication for token introspection).
Output schemas not documented. LLMs cannot see what fields verify_agent, get_popa, get_popa_leaderboard, and lookup_audit_token return. This forces agents to guess at response structure and risks failed downstream tool chaining.
No error recovery guidance. Tools lack descriptions of failure modes and what LLMs should do next (e.g., 'If agent_id not found, try get_popa_leaderboard to discover valid agents'). Errors will be opaque to the agent.
get_popa_leaderboard lacks pagination parameters. No limit default stated in description (only in schema). No next_cursor or total_count documented. Returning 100 agents could blow context; LLMs need explicit pagination guidance.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
Verify an AgentLair agent's behavioral trust score. Returns the full trust profile: overall score [0, 100], confidence interval, ATF maturity level (intern → junior → senior → principal), and per-dimension breakdown (consistency, restraint, transparency). Use this when you need to know whether to trust another agent before delegating, paying, or extending capability. The score is computed from the agent's audit trail — behavior, not declarations. Anonymous callers pay 0.01 USDC via x402. Set AGENTLAIR_AAT to skip payment.
Tool annotations missing. No readOnlyHint, destructiveHint, or idempotentHint declared. All 5 tools are read-only (safe to retry), but this is not signaled to the MCP client or LLM.
STDIO transport only. Server is not remotely accessible and cannot be used by hosted MCP clients (e.g., Claude web, Cursor cloud). Limits deployment to local/desktop environments.