aiden-mcp
Official# aiden-mcp
[MCP](https://modelcontextprotocol.io) server for **AIDEN**, the identity and reputation layer for AI agents on
[Robinhood Chain](https://docs.robinhood.com/chain/). It lets Claude and other MCP clients look up agents registered in the
**ERC-8004** registries: profiles, owners, feedback, validations and USDG-weighted trust scores.
Read-only. No wallet or API key needed.
## Tools
| Tool | What it does |
|---|---|
| `search_agents` | Full-text search, owner filter, sort by `newest` / `score` / `feedback` / `settled` / `relevance` |
| `get_agent` | Full record: profile, reputation, validations, trust score, settlement evidence |
| `get_agent_feedback` | Reputation Registry feedback, newest first, classified and normalised |
| `get_agent_validations` | Latest 100 Validation Registry entries |
| `get_registry_stats` | Index totals |
| `get_indexer_health` | How far the indexer is behind the chain head |
## Connect
### Remote (claude.ai, ChatGPT, any URL-based client)
```
https://mcp.aiden.network/mcp
```
Streamable HTTP, stateless, no authentication.
- **claude.ai / Claude Desktop:** Settings → Connectors → Add custom connector → paste the URL.
- **ChatGPT:** Settings → Apps & Connectors → Advanced → enable Developer mode → Create → paste the URL, authentication: none.
- **Claude Code:** `claude mcp add --transport http aiden https://mcp.aiden.network/mcp`
### Local (stdio via npx)
No install step needed; clients run it with `npx`.
### Claude Code
```bash
claude mcp add aiden -- npx -y aiden-mcp
```
### Claude Desktop / Cursor / other clients
```json
{
"mcpServers": {
"aiden": {
"command": "npx",
"args": ["-y", "aiden-mcp"]
}
}
}
```
## Configuration
| Variable | Default | |
|---|---|---|
| `AIDEN_API_URL` | `https://api.aiden.network` | AIDEN indexer API. Use `http://127.0.0.1:4000` for a local indexer |
| `AIDEN_TIMEOUT_MS` | `10000` | Per-request timeout, 1000 to 60000 |
## Development
```bash
git clone https://github.com/aiden-network/aiden-mcp.git && cd aiden-mcp && npm install
npm run dev # run from source (tsx)
npm test # vitest
npm run test:coverage # 80% threshold enforced
npm run inspect # MCP Inspector against dist/
npm run worker:dev # remote endpoint locally (wrangler)
npm run worker:deploy # typecheck, test, deploy the Worker
```
The remote endpoint is `worker/index.ts`: a Cloudflare Worker that serves the same tools over Streamable HTTP
(`src/http.ts`). It reaches the AIDEN API through a service binding and forwards the caller's IP, so API rate
limits apply per user.
## License
MIT
TDQS
Scored across 6 tools
Most tools target clearly distinct resources: feedback, validations, agent detail, search, stats, and indexer health. The only mild overlap is that get_agent returns a reputation/validation summary while get_agent_feedback and get_agent_validations return detailed registry entries, so an agent might occasionally pick the summary tool when it needs the full detail.
All tools use a consistent snake_case verb_noun pattern: get_agent_feedback, get_agent_validations, get_indexer_health, search_agents, get_agent, get_registry_stats. The verb choice (get for retrieval, search for discovery) is predictable and uniform.
Six tools is well-scoped for a read-only agent directory and registry explorer. Each tool covers a distinct facet (search, detail, reputation, validation, aggregate stats, indexer health) without redundancy.
Discovery, detail, reputation, validation, aggregate stats, and indexer health are all covered, which is strong for this domain. Minor gaps include no dedicated owner-centric lookup or explicit pagination/listing tool, though search_agents with owner filtering partially compensates.