BCA Crypto Brain
The editorial ground truth for crypto LLM agents. An open Model Context Protocol server that gives Claude, ChatGPT, LangChain, Eliza, and Virtuals Protocol agents dated, cited, cross-referenced crypto intelligence — instead of hallucinated facts from pretraining.
Published as @blockchainacademics/mcp on npm and bca-mcp on PyPI. MIT licensed. One command to install.
Start here
@blockchainacademics/mcp into Claude Desktop in 60 seconds. Also works in Cursor, Windsurf, and any stdio-speaking client.Install the MCPFree tier is 2,000 calls/month, no card required. Paid tiers unlock agent-backed generators and higher rate limits.Get an API keyAll 99 tools, grouped by category, with input schemas, tier badges, and response envelope examples. Auto-generated from the Zod schemas.Browse the tool referenceWhat you get
- 99 typed tools across news, market data, on-chain analytics, sentiment, directories, fundamentals, events, dev infrastructure, portfolio, microstructure, regulatory, security, narrative, chains, memes, compute, and services.
- 6 proprietary indicator composites — Coverage Index, Narrative Pulse, Dev Activity, Sentiment Momentum, Regulatory Risk, Macro Exposure — scored daily across 500+ assets.
- 5 agent-backed generators —
due_diligence,tokenomics_model,summarize_whitepaper,translate_contract,monitor_keyword. Long-running jobs that spin off Groq-powered workers and stream back polished markdown. - Attribution envelope on every response — every tool call returns
{data, attribution: {citations: [...]}, meta}. Each citation carriescite_url,as_of, andsource_hash. Your agent never has to guess where a fact came from, and your users can always click through. See Response envelope.
Why BCA over pretraining?
Pretraining is a frozen snapshot. BCA is live editorial: ~10 articles/day, 3,500+ archived pieces, an entity graph with ~6,000 canonical crypto projects, tokens, people, and institutions — all reconciled to a single source of truth, all timestamped, all citable.
When an agent calls search_news("circle IPO"), it gets titles + summaries + cite_url links that resolve to published articles on blockchainacademics.com — each carrying UTM params so we attribute the agent caller correctly. No hallucinated dates, no made-up quotes, no stale pretraining.
Supported clients
| Client | Transport | Status |
|---|---|---|
| Claude Desktop | stdio | Supported (v0.2) |
| LangChain (Python) | stdio subprocess | Supported (v0.2) |
| Cursor / Windsurf | stdio | Supported (v0.2) |
| Eliza | plugin | In development (v0.3) |
| Virtuals Protocol | HTTP / SSE | In development (v0.3) |
| Custom | stdio or HTTP shim | Supported (v0.2) |
Pricing & attribution
Free and Starter tiers include every tool, rate-limited. Pro and above unlock agent-job skills, webhook delivery, and strip_attribution opt-out. Full tier matrix on Tiers & quotas.
Attribution is required on Free and Starter. See Attribution & licensing for the citation contract and
llms.txtreference.
Need help?
- GitHub:
blockchainacademics/bca-mcp— issues, PRs, roadmap - Status:
status.blockchainacademics.com(arrives v0.3) - Email:
dev@blockchainacademics.com