generated: '2026-09-19'
method: probed
status: published
source: https://prediction.mcp.chainaware.ai/sse
docs: https://chainaware.ai/learn/prediction-mcp/index.html
setup_docs: https://chainaware.ai/learn/prediction-mcp/setup.html
repository: https://github.com/ChainAware/behavioral-prediction-mcp
registry:
  official_mcp_registry:
    name: io.github.ChainAware/chainaware-behavioral-prediction-mcp
    version: 1.0.0
    published: '2026-01-19'
    status: active
    remote: {type: sse, url: 'https://prediction.mcp.chainaware.ai/sse', header: X-API-Key (isRequired true per server.json)}
    url: https://registry.modelcontextprotocol.io/v0/servers?search=chainaware
  glama: {name: behavioral-prediction-mcp, framework: FastMCP 2.12.0, transport: [sse]}
summary: >-
  ChainAware operates ONE hosted MCP server, reachable at two hostnames that serve the same 25,293-byte landing
  page and the same tool set: prediction.mcp.chainaware.ai (the name every provider document uses — docs,
  README, server.json, the MCP registry, AGENTS.md, SKILL.md) and mcp.chainaware.ai (the name in llms.txt and
  the site sitemap). Transport is legacy HTTP+SSE: GET /sse opens the stream and emits an endpoint event
  naming POST /messages/?session_id=<uuid>; each POST is acknowledged 202 Accepted and the JSON-RPC reply
  arrives on the stream. initialize and tools/list answered ANONYMOUSLY on 2026-09-19 — serverInfo
  {name: ChainAwareMCP, version: 1.11.0}, protocolVersion 2025-03-26, capabilities tools/prompts/resources
  (listChanged false), no instructions — returning fourteen tools with real inputSchemas (saved verbatim to
  mcp/chainaware-ai-mcp-tools.json). Authorship is unambiguous: the tool names are ChainAware products, the
  descriptions cite ChainAware accuracy claims, and the MIT-licensed FastMCP source is on the provider's
  GitHub. The declared MCP paths /mcp (Streamable HTTP) and /.well-known/oauth-protected-resource both 404 on
  both hosts; the only well-known documents the MCP hosts serve are /.well-known/x402 and /.well-known/x402.json.
deployment:
  mode: remote
  endpoint: https://prediction.mcp.chainaware.ai/sse
  alias_endpoint: https://mcp.chainaware.ai/sse
  transport: sse
  messages_endpoint: https://prediction.mcp.chainaware.ai/messages/?session_id=<from the SSE endpoint event>
  auth: api-key
  verified: probed
  note: >-
    A hosted HTTPS SSE endpoint an MCP client connects to directly. No stdio package is published on npm or
    PyPI; the documented Claude Desktop config bridges the remote server with `npx mcp-remote <sse url>
    --header "X-API-Key: ..."`, which is a client bridge, not a local server. The FastMCP source can be run
    locally (the README's example clients point at http://localhost:5000/sse) but the provider documents no
    install command for that, so mode is remote, not both. auth is api-key in three documented forms: an
    X-API-Key header on the SSE connection (Claude Code / Cursor / SDK examples), an ?apiKey= query parameter
    on the SSE URL (ChatGPT and Claude Web integrations), and — for the six tools that require it — an
    in-band `apiKey` argument inside the tool call. The connection itself, initialize and tools/list need no
    credential. Eight tools need no key at all per the provider's docs and the live schemas (token_rank_list,
    token_rank_single, run_token_audit, get_token_audit_result, agents_trust_score_list,
    agents_trust_score_single, check_job_status, get_job_results). x402 pay-per-call is declared "at the
    backend API level" (/.well-known/x402.json: backend https://api.chainaware.ai) and the README says the
    apiKey argument may be omitted when paying with x402; no OAuth metadata is served on any host.
servers:
- id: chainaware-behavioral-prediction
  name: ChainAwareMCP
  endpoint: https://prediction.mcp.chainaware.ai/sse
  transport: sse
  auth: api-key (X-API-Key header, ?apiKey= query, or in-band apiKey tool argument); anonymous initialize/tools/list
  payment: x402 (USDC on Base) supported at the backing API; HTTP 402 challenge documented in the README tool error tables
  status: live
  probe:
    fetched: '2026-09-19'
    sse_open: {http_status: 200, content_type: 'text/event-stream; charset=utf-8', server: uvicorn, via: CloudFront}
    endpoint_event: /messages/?session_id=<uuid>
    initialize:
      post_status: 202
      protocol_version: '2025-03-26'
      server_info: {name: ChainAwareMCP, version: 1.11.0}
      capabilities: {experimental: {}, prompts: {listChanged: false}, resources: {subscribe: false, listChanged: false}, tools: {listChanged: false}}
    tools_list:
      post_status: 202
      tool_count: 14
      tools_with_output_schema: 10
      file: mcp/chainaware-ai-mcp-tools.json
    other_paths:
      - {url: 'https://mcp.chainaware.ai/mcp', status: 404, note: no Streamable HTTP endpoint}
      - {url: 'https://mcp.chainaware.ai/.well-known/oauth-protected-resource', status: 404}
      - {url: 'https://prediction.mcp.chainaware.ai/.well-known/oauth-protected-resource', status: 404}
      - {url: 'https://mcp.chainaware.ai/.well-known/x402.json', status: 200, file: well-known/chainaware-ai-x402.json}
    landing_page_endpoints:
      - 'GET / — landing page listing the 14 tools'
      - 'GET /sse — MCP SSE stream'
      - 'POST /messages/ — MCP message endpoint keyed by session_id'
      - 'POST /chat — single-turn GPT-4o agent that picks and calls a tool (non-MCP convenience endpoint)'
      - 'POST /generate-stream — streaming Qwen3-32b agent response (non-MCP convenience endpoint)'
      - 'GET /.well-known/x402 and /.well-known/x402.json — x402 payment discovery'
version_drift:
  note: >-
    Three version numbers coexist: the live server reports 1.11.0; server.json and the MCP registry entry
    (published 2026-01-19) say 1.0.0; the repository's last commit is 2026-08-11. The provider's own pages also
    disagree on tool count — the landing page and llms.txt say 14 (correct), the docs nav says 6, the setup
    page says "all ten tools", and the subagent count is given as 32 or 34 (the repo ships 34 files).
tool_count: 14
tools:
- {name: predictive_fraud, category: fraud, key_required: true, description: 'AI fraud probability score (0-1) and AML forensic flags for a single wallet. Networks ETH, BNB, POLYGON, TON, BASE, TRON, HAQQ.', rest_operation: checkWalletFraud}
- {name: predictive_fraud_batch, category: fraud-batch, key_required: true, description: 'Schedule fraud analysis for up to 1,000 wallets; returns job_id + signature immediately.', rest_operation: null}
- {name: predictive_behaviour, category: behaviour, key_required: true, description: 'Next-action predictions, risk profile, DeFi category segmentation and personalised recommendations for a wallet. Networks ETH, BNB, BASE, HAQQ, SOLANA.', rest_operation: auditWalletBehaviour}
- {name: predictive_behaviour_batch, category: behaviour-batch, key_required: true, description: 'Schedule behaviour analysis for up to 1,000 wallets; returns job_id + signature immediately.', rest_operation: null}
- {name: predictive_rug_pull, category: rug-pull, key_required: true, description: 'Forecasts whether a liquidity pool or contract is likely to rug pull; returns risk indicators and liquidity events. Networks ETH, BNB, BASE, HAQQ.', rest_operation: checkRugPull}
- {name: credit_score, category: credit, key_required: true, description: 'AI-driven crypto trust score (1-9) combining inflows/outflows, fraud signals and social graph. ETH only.', rest_operation: getCreditScore}
- {name: token_rank_list, category: token-rank, key_required: false, description: 'Paginated, filterable list of tokens ranked by holder community strength (limit/offset, sort, category filter, name search). Networks ETH, BNB, BASE, SOLANA.', rest_operation: null}
- {name: token_rank_single, category: token-rank, key_required: false, description: 'Community rank and top holders for a single token contract address and network.', rest_operation: null}
- {name: run_token_audit, category: token-audit, key_required: false, description: 'Get-or-create: returns a cached full audit immediately, or queues a new one returning a job_id. Lowercase network ids eth, bsc, base, arbitrum, avalanche, optimism, polygon.', rest_operation: null}
- {name: get_token_audit_result, category: token-audit, key_required: false, description: 'Poll for a completed token audit; full risk report when audit_status = "complete".', rest_operation: null}
- {name: agents_trust_score_list, category: agent-trust, key_required: false, description: 'Paginated list of ERC-8004 AI agents with 0-1000 on-chain trust scores and tiers (page/limit, sort_by, sort_order, registered_after).', rest_operation: null}
- {name: agents_trust_score_single, category: agent-trust, key_required: false, description: 'Detailed trust report for a single ERC-8004 agent by agent_id and chain_id.', rest_operation: null}
- {name: check_job_status, category: batch-jobs, key_required: false, description: 'Progress of a batch job (completed / failed / pending counts). Requires job_id + signature.', rest_operation: null}
- {name: get_job_results, category: batch-jobs, key_required: false, description: 'Wallet address list from a completed or partial batch job. Call only when status is completed or partial.', rest_operation: null}
subagents:
  count: 34
  location: https://github.com/ChainAware/behavioral-prediction-mcp/tree/main/.claude/agents
  index: skills/chainaware-ai-AGENTS.md
  note: Provider-published Claude Code subagent definitions wrapping these tools; also mirrored as Eliza character files (eliza/characters/) and an OpenAI skill definition (agents/openai.yaml) in the same repository.