dossier-mcp 0.1.0

Dossier MCP server — onchain risk scoring for Base (token / deployer / wallet), callable by AI agents.

Dossier MCP server

Onchain risk scoring for Base, callable by AI agents (Claude, Cursor, …) over the Model Context Protocol. A thin client over the Dossier REST API.

Tools

Tool	What it answers	Score direction
dossier_scan_token	Is this token a honeypot / rug / high-risk?	risk — higher = worse
dossier_scan_deployer	Can I trust the dev who shipped it?	trust — higher = better
dossier_scan_portfolio	How exposed is this wallet (holdings + approvals)?	risk — higher = worse

Risk and trust are kept as separate scores with explicit direction — never collapsed into one number.

Configuration (env)

Var	Default	Purpose
DOSSIER_API_BASE	http://localhost:8000	Base URL of the Dossier REST API
DOSSIER_MCP_TRANSPORT	stdio	stdio (local) or http (streamable-HTTP, remote)
DOSSIER_MCP_HOST / DOSSIER_MCP_PORT	127.0.0.1 / 8765	bind address for HTTP transport

Use it from Claude Desktop / Cursor (stdio)

{
  "mcpServers": {
    "dossier": {
      "command": "uvx",
      "args": ["dossier-mcp"],
      "env": { "DOSSIER_API_BASE": "https://your-dossier-host" }
    }
  }
}

Or from a local checkout: uv run dossier-mcp (with DOSSIER_API_BASE pointing at your backend).

Run as a hosted HTTP server (multi-client)

DOSSIER_MCP_TRANSPORT=http DOSSIER_MCP_PORT=8765 DOSSIER_API_BASE=http://localhost:8000 uv run dossier-mcp

Agents then connect to the streamable-HTTP endpoint at http://host:8765/mcp.

Example

"Is 0x833589fcd6edb6e08f4c7c32d4f71b54bda02913 safe to buy on Base?"

The agent calls dossier_scan_token → "USDC — risk: low (29/100, higher = worse; 60% confidence)…", then can follow up with dossier_scan_deployer on the attributed creator.

Dossier provides risk intelligence, not financial advice; false positives and negatives are possible.