AgentScore MCP Server
# AgentScore MCP Server
Official MCP server for the AgentScore Registry (Base Sepolia).
## Quick Start
```bash
npm install
npm run build
npm run dev # stdio (Claude Desktop, Cursor, etc.)
# or
HTTP_PORT=3000 npm start # HTTP/SSE
```
## Docker (recommended for HTTP)
```bash
npm run docker:build
npm run docker:run
# Server available at http://localhost:3000/mcp
```
## Configure in MCP Clients
Claude Desktop / Cursor / Windsurf
Add to your MCP config (usually `~/Library/Application Support/Claude/mcp.json` or equivalent):
```json
{
"mcpServers": {
"agent-score": {
"command": "node",
"args": ["/absolute/path/to/agent-score-mcp/dist/index.js"]
}
}
}
```
### For Docker version:
```json
{
"mcpServers": {
"agent-score": {
"command": "docker",
"args": ["run", "-i", "--rm", "agent-score-mcp"]
}
}
}
```
## HTTP clients (e.g. custom agents)
POST to `http://localhost:3000/mcp` with standard MCP JSON-RPC.
### Available Tools
- `getAgentScore(identifier)` → `{score, historyCount, metadataURI}`
- `getAgentHistory(tokenId, limit)` → `[{score, timestamp}, ...]`
- `listRegisteredAgents(owner?)` → array of enriched agents
All read-only. Zero private keys. Fully trustless.
Enjoy building with AgentScore reputation! 🚀
TDQS
Scored across 3 tools
Each tool has a clearly distinct purpose: getAgentHistory retrieves historical data, getAgentScore provides current reputation metrics, and listRegisteredAgents enumerates agents with optional filtering. There is no overlap in functionality, and the descriptions clearly differentiate their roles.
The naming is mixed: getAgentHistory and getAgentScore follow a consistent verb_noun pattern, but listRegisteredAgents uses a different verb style ('list' vs. 'get'). While all names are readable, the inconsistency in verb choice reduces predictability.
With only 3 tools, the set feels thin for a server named 'AgentScore MCP Server', which suggests a broader scope for managing agent scores and data. While the tools cover basic retrieval and listing, the count may be insufficient for comprehensive operations like updating scores or managing registrations.
The tool surface is significantly incomplete for an agent scoring system. It only includes read operations (get and list), with no ability to create, update, or delete scores or agents. This leaves obvious gaps in CRUD coverage, likely causing agent failures in scenarios requiring data modification.