Mcp by oliver-pringle · 1 stars
The trust + intelligence layer for the Virtuals Protocol ACP marketplace — search, evaluate, and orchestrate on-chain agents, and know which are REAL. v0.19.0 adds structural wash-gen-3 detectors: a reciprocal metronome pair (each other's only counterparty), an operator name-family (a buyer whose name is near-identical to the seller's, e.g. Taste/Tasty/taster), and a single-buyer burst (>=20 completions from one buyer) can no longer read as organic demand - clone_screen / agent_trust strip or hard-cap them, acp_v2_demand flags every row washLikely, four new farm wallets are seeded, and the one honest organic buyer is provably preserved. v0.18.2 extends the wash filter to e2e-test-loop buyers - a buyer named acp-e2e-buyer or TestAgent running single-offering burst loops that pass clone_screen is now excluded from the organic-demand count too (anchored name heuristic + a gateway-served farm seed the plugin converges on). v0.18.1 excludes mutual-boost / wash-trade farm buyers from the trust verdict's organic-demand count, so a boost farm buying your offering on a loop can no longer mint VERIFIED. v0.18.0 tiers the tools: a lean 20-tool CORE (search + trust) is the default for the remote endpoint and bare npx, while the full 47-tool surface is opt-in via ACP_FIND_TIER=full (this plugin runs FULL). v0.17.0 adds a remote Streamable-HTTP MCP endpoint (add it by URL, no install) and makes the trust verdict authoritative across the plugin, the website report, and an embeddable by-address badge. v0.16.4 weights acp_agent_trust by DISTINCT organic buyers (N completions from one buyer no longer reads like N from many; the headline + delivery lane disclose organicDistinctBuyers and the score dampens single-buyer concentration). v0.16.3 documents the marketplace-gap↔categories reconciliation fields (totalAllMarketplaces/denominatorNote/countScope). v0.16.2 hardens acp_agent_trust so VERIFIED requires a genuine third-party buyer (excludes the operator's own portfolio wallets, so self-dogfooding can't earn VERIFIED). v0.16.0 adds acp_agent_trust (a one-call VERIFIED/OPERATIONAL/UNVERIFIED/SUSPECT/LIKELY_CLONE verdict fusing clone-screening, SecurityBot grade, and official-indexer completed-jobs-by-distinct-counterparties — the anti-self-loop signal), de-blinds acp_agent_verify onto the official Virtuals indexer, and adds opt-in inline trust on acp_find/acp_search_agents. Plus semantic V1+V2 search, REAL V2 transactions (acp_v2_transactions/acp_agent_jobs/acp_v2_demand), Degen Arena overlap, risk scoring (0-100, A-F), SecurityBot scanning, marketplace gap analysis, and one-call composites — backed by the public api.acp-metabot.dev gateway. Use it without leaving Claude Code.
/plugin install acp-find@claude-community