rwa-attest

Cryptographically verifiable RWA (Real-World Asset) trust attestation + multi-chain DeFi data + a free protocol positioning scorecard. Exposed via Model Context Protocol so any MCP-capable AI assistant (Claude Desktop, Cursor, Cline, etc.) can call it.

8 signed RWA attestations live with real-fetched proofs: BlackRock BUIDL, Ondo OUSG, Ondo USDY, Maple Finance, Centrifuge Protocol, Hashnote USYC, Superstate, Spiko. Bytecode + SHA256 from Ethereum RPC, ERC20 metadata via eth_call, live USD prices via CoinGecko, audit URL liveness via HTTPS HEAD/GET. Every signature independently verifiable from the embedded ed25519 public keys + canonical JSON.

Built by Tashoma Vilini — DeFi PMM, Liqwid Finance ($81M TVL peak on Cardano).

What it does

Ten tools, all read-only:

Data source for chain/protocol tools: DefiLlama public API. No paid keys required for the data tools.

Related MCP server: mcp-server-insumer

Why use it

• Built for where RWAs actually live. Most RWA TVL ($15B+) is on Ethereum — BlackRock BUIDL, Ondo OUSG/USDY, Maple, Centrifuge, Hashnote USYC, Superstate, Spiko. This server is RWA-first and Ethereum-default.

• Cardano-aware lens. Author was a Cardano DeFi operator (Liqwid). Cardano's $130M DeFi ecosystem gets equal treatment — including the small but growing RWA presence (KAIO, Mehen).

• Free positioning scorecard. Tell it a protocol name; get a 5-dimension teardown. If the surface-level diagnosis is useful, the upsell path is a paid full teardown — DM @defibabylon on X.

• Verifiable attestations, not vibes. The RWA attestation tool returns ed25519-signed trust evidence with reproducible canonical JSON. Most DeFi "trust scores" are opaque dashboards. Ours are cryptographically verifiable from the public keys + the policy + the inputs.

Security model

Built defense-in-depth, by construction, not by convention. Every tool routes through:

What this server CANNOT do, by design

• Cannot execute arbitrary code

• Cannot read your filesystem beyond the 11 hard-coded files above

• Cannot make HTTP requests to any host outside the 5-entry allowlist

• Cannot write any file

• Cannot start any subprocess

• Cannot exfiltrate environment variables, .env contents, or API keys

• Cannot leak filesystem paths under /root/.hermes/keys/, /root/Obsidian Vault/, /root/CVs/, /root/.claude/projects/-root/memory/, or any operator-personal location

If the server detects a non-allowlisted host, a non-allowlisted file read, or a response that would contain secret-shaped data, it refuses with a generic error and logs the violation server-side. No exception messages, file paths, or stack traces are surfaced to the caller.

Install (Claude Desktop)

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (or platform equivalent):

{
  "mcpServers": {
    "rwa-attest": {
      "command": "python",
      "args": ["-m", "rwa_attest.server"],
      "env": {
        "ANTHROPIC_API_KEY_FOR_MCP": "<your-anthropic-key-for-scorecard-tool>"
      }
    }
  }
}

The ANTHROPIC_API_KEY_FOR_MCP env var is optional — only needed if you want the score_protocol_positioning tool to actually score (it falls back to returning raw protocol metadata otherwise). The other 9 tools work without any API key.

Install (Smithery)

npx @smithery/cli install rwa-attest --client claude

(once published — see smithery.yaml)

Run locally

python -m rwa_attest.server

Speaks MCP via stdio. Compatible with any MCP client.

Example queries

Once installed, ask your AI assistant things like:

• "What's the total RWA TVL right now and which chain dominates?"

• "List the top 5 Ethereum RWA protocols by TVL."

• "Compare BlackRock BUIDL and Ondo Yield Assets — which has stronger 7d momentum?"

• "Score Aave V3's positioning on the 5 dimensions."

• "Show me the most recent signed RWA trust attestation."

• "Give me Cardano's DeFi status." — Cardano is here too, just not the default.

Architecture

rwa_attest/
├── security.py                # Allowlists, validators, secrets scanner, rate limiter
├── server.py                  # MCP entry point, tool registry, dispatch
└── tools/
    ├── chain.py               # Multi-chain DeFi data (DefiLlama) — Ethereum default
    ├── rwa.py                 # RWA landscape + top protocols + protocol detail
    ├── positioning.py         # FREE 5-dimension scorecard (sandboxed LLM)
    └── attestation.py         # Wraps the signed RWA attestation engine output

Every tool routes through:

• validate_*(arg) for inputs

• safe_get(url) for outbound HTTP (host allowlist)

• safe_read(path) for file reads (path allowlist)

• safe_response(payload) for outbound responses (secrets scanner)

This is enforced at the function level. There is no fall-through path that bypasses the security layer.

License

MIT.

Author

Tashoma Vilini — DeFi PMM, Liqwid Finance ($81M TVL peak on Cardano).

For a full DeFi positioning teardown, GTM sprint, or custom RWA trust attestation: DM @defibabylon on X.