Wicked MCP README

WickedAPI MCP Server

MCP server for the whole Wicked suite — trading data, DeFi protocol health, token safety, brokerage sync, agent identity, reputation, a tool-reliability registry, hallucination checks, persistent agent memory, and x402 paywalls, in one server.

90 tools across 12 live services. The first four are generated automatically from each service's own live openapi.json, so the list always matches what's deployed; the rest are hand-written because they use wallet signatures and non-/v1 routes:

Plus 6 resources (the platform directory, llms.txt, and each service's live endpoint catalog) and 7 prompts (rug_check, vet_dependency, morning_briefing, fact_check, verify_agent, find_reliable_tool, agent_memory_guide) that chain tools together the way the cookbook does.

These tools never sign anything or hold a key. Agents with no wallet can still try Wicked Identity end to end through a sandbox identity (identity_sandbox_register, then pass sandbox_token instead of a signature); sandbox results are for testing and are never reported as verified. Identity (real wallet), Registry registration and Memory take a signature you produce with your own wallet (identity_get_nonce / memory_prepare return the exact message to sign). Calls that need x402 payment return the real 402 challenge (decoded under payment_required for x402 v2) rather than auto-paying — WICKEDAPI_PAYER_PRIVATE_KEY auto-pay applies to the Trading, Protocol Health, Token Risk and Broker Sync tools.

Remote endpoint (no install)

The same 90 tools are served over streamable HTTP at https://mcp.wickedapi.com/mcp (stateless, rate-limited per IP; the public deployment carries no personal keys, so paid calls return the x402 challenge). Point any streamable-HTTP MCP client at it. To self-host, build the Dockerfile (or npm run build && npm run start:http); see .env.example for RATE_LIMIT_PER_MINUTE, ALLOWED_HOSTS and TRUSTED_CLIENT_IP_HEADER.

No signup required

Every tool hits a real endpoint gated by x402 (HTTP-native pay-per-call, USDC on Base) with an API-key fallback — and the Token Risk Oracle tool is free no matter what:

• Zero config — calls without credentials still work; paid endpoints return the real 402 payment challenge (price, network, payTo) as the tool result.
• Autonomous pay-per-call — set WICKEDAPI_PAYER_PRIVATE_KEY to a funded Base-mainnet wallet's private key, and every call auto-pays ~$0.001 in USDC and returns real data. No account, no subscription.
• Partner API key — set WICKEDAPI_API_KEY for free, rate-limited access across all four services (same key works everywhere).

Install

{
  "mcpServers": {
    "wickedapi": {
      "command": "npx",
      "args": ["-y", "@wickedlabs/wicked-mcp"],
      "env": {
        "WICKEDAPI_API_KEY": "your-key-here"
      }
    }
  }
}

Or for autonomous pay-per-call instead of a key:

{
  "mcpServers": {
    "wickedapi": {
      "command": "npx",
      "args": ["-y", "@wickedlabs/wicked-mcp"],
      "env": {
        "WICKEDAPI_PAYER_PRIVATE_KEY": "0xyour-base-wallet-private-key"
      }
    }
  }
}

Nothing configured? The Token Risk Oracle tool and the 5 free Trading Data API endpoints (sentiment, macro_calendar, market_overview, gas, kalshi_probability) still work with zero setup.

Resources

Read without calling a tool — useful for a client that surfaces resources in its UI:

• wickedapi://platform/services — the full service directory (JSON)
• wickedapi://platform/llms — the platform's llms.txt
• wickedapi://trading/status, wickedapi://protocol_health/status, wickedapi://token_risk/status, wickedapi://broker_sync/status — each service's live endpoint catalog

Quick start

claude mcp add --transport http wicked https://mcp.wickedapi.com/mcp   # Claude Code, no install
npx -y @wickedlabs/wicked-mcp                                                    # local stdio

Claude Desktop / Cursor / VS Code config and per-framework starters: see the cookbooks. The remote also serves a landing page at /, /llms.txt, an MCP server card at /.well-known/mcp/server-card.json, and aggregate per-tool call counts at /stats (counts only; no arguments or IPs).

Prompts

One-click workflows that chain real tool calls (Claude Desktop and other prompt-aware clients surface these directly):

• rug_check (address, chain, optional symbol) — token risk check + market sentiment + optional momentum read, before buying.
• vet_dependency (slug, optional token_address, chain) — protocol health + optional contract-level risk check, before building on a DeFi protocol.
• morning_briefing (optional symbol) — macro calendar + earnings calendar + sentiment + momentum verdict, as one digest.
• fact_check (output, optional source) — split an answer into claims and verify each with Wicked Sanity.
• verify_agent (wallet) — Identity assertion + Reputation stake/tier before you transact with an agent.
• find_reliable_tool (need, optional category, min_score) — Registry search by real uptime/latency/schema scores.
• agent_memory_guide (wallet, task) — the two-step signed Memory flow, explained to the model.

Environment variables

If neither WICKEDAPI_API_KEY nor WICKEDAPI_PAYER_PRIVATE_KEY is set, paid calls still succeed at the protocol level — you'll get the real 402 challenge back as the tool result instead of data. If one service's spec fails to load (network hiccup, that service is down), its tools are simply skipped for the session — the rest still register normally.

Local development

npm install
npm test            # builds, then unit + end-to-end tests (e2e needs network)
node dist/index.js  # stdio
node dist/http.js   # streamable HTTP on $PORT (default 8080)

What this is not

This server doesn't reimplement any data logic — it's a thin bridge from MCP tool calls to the live Wicked HTTP endpoints across all ten services. All the actual market data, indicators, scoring, and risk signals live in the APIs themselves.