MCP server

Ethos serves its most useful reputation lookups as an MCP server, so an agent that installs tools gets them without reading the API reference or paying over x402:

https://developers.ethos.network/mcp

Streamable HTTP, no authentication, read-only. Five tools, one per lookup, each answering with the same JSON the free API returns.

Tool	Takes	Returns	Same as
`ethos_score`	`userkey`	Credibility score (0 to 2800) and level	`GET /api/v2/score/userkey`
`ethos_user_by_x`	`handle` (X handle or account id)	The Ethos profile behind an X account, every linked account	`GET /api/v2/user/by/x/{accountIdOrUsername}`
`ethos_user_by_address`	`address` (EVM, `0x…`)	The Ethos profile holding a wallet	`GET /api/v2/user/by/address/{address}`
`ethos_search_users`	`query`, optional `limit`, `offset`	Profiles matching a name, handle or address fragment	`GET /api/v2/users/search`
`ethos_xp`	`userkey`	Total XP across all seasons	`GET /api/v2/xp/user/{userkey}`

A userkey names a person any way Ethos knows them: address:0x…, profileId:1, service:x.com:username:ethos_network, service:x.com:<account id>, service:discord:<id> or service:telegram:<id>.

The server’s initialize answer carries instructions on which tool to reach for and how to read a score (below 800 untrusted, 800 to 1199 questionable, 1200 to 1399 neutral, 1400 and up increasingly reputable). Scores reflect peer judgement and can be gamed at the margins: a strong signal, not a verdict.

Connect a client

claude mcp add --transport http ethos https://developers.ethos.network/mcp

In .cursor/mcp.json (per project) or ~/.cursor/mcp.json (everywhere):

{
  "mcpServers": {
    "ethos": { "url": "https://developers.ethos.network/mcp" }
  }
}

Settings → Connectors → Add custom connector: name Ethos, URL https://developers.ethos.network/mcp, no OAuth. For a Claude Desktop that only takes local servers, claude_desktop_config.json can bridge to it:

{
  "mcpServers": {
    "ethos": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://developers.ethos.network/mcp"]
    }
  }
}

The endpoint is plain JSON-RPC over POST, so a shell can list and call the tools:

curl -s https://developers.ethos.network/mcp \
  -H "Content-Type: application/json" -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

curl -s https://developers.ethos.network/mcp \
  -H "Content-Type: application/json" -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"ethos_score","arguments":{"userkey":"service:x.com:username:ethos_network"}}}'
# {"result":{"content":[{"type":"text","text":"{\"score\":2441,\"level\":\"revered\"}"}]},"jsonrpc":"2.0","id":2}

Any client that speaks MCP’s streamable HTTP transport works the same way; a GET on the URL from a browser describes the endpoint.

Details

Stateless. The server issues no session id and keeps nothing between requests, so a client can send tools/list or tools/call without an initialize first. Responses are JSON, not server-sent events.
Errors. A lookup the API refuses (an unknown user, a malformed address) comes back as a tool error carrying the API’s status and message, not as a protocol failure.
Rate limits are the free API’s. Each call reaches api.ethos.network with the client header developers.ethos.network-mcp.
Only reads. Nothing here writes. For reviews, vouches or XP tips an agent uses the full API with an API key.
Listed in the MCP registry as network.ethos/api, where MCP clients and directories that read it find it.

Need more than these five? The whole API is free with an X-Ethos-Client header; the Vibe Coding Quickstart is written for an agent to read, and Agents and x402 covers the paid-per-request copies of the same lookups.