本文へスキップ

Hermes AI API MCP server

A remote MCP server any client can reach over HTTP. Add one URL, sign in with your Hermes AI account, and your agent can find models, run tasks and fetch results, paid from your prepaid credits.

https://api.hermes-ai.net/mcp

エンドポイント

This URL is the server itself, not a web page. MCP clients connect to it; opened in a browser it answers 401 and asks for sign-in.

Setting値
エンドポイントhttps://api.hermes-ai.net/mcp
TransportStreamable HTTP
Server namehermes-ai-api
Sign-inOAuth 2.1 with PKCE and dynamic client registration
Without OAuthAuthorization: Bearer <API key>
Resource metadatahttps://api.hermes-ai.net/.well-known/oauth-protected-resource/mcp
Authorization serverhttps://api.hermes-ai.net/.well-known/oauth-authorization-server/api/auth

Add it to your client

Pick your client and copy the command or config. Each one has a full guide in the docs.

terminal
claude mcp add --transport http hermes-ai-api https://api.hermes-ai.net/mcp
Claude Code setup guide

How sign-in works

You sign in once in your browser. No API key is copied into the client.

  1. 1Your client calls the endpoint without a token. The server answers 401 with a WWW-Authenticate header that points to its resource metadata.
  2. 2The client reads that metadata, finds the authorization server and registers itself. Registration is open and requires PKCE.
  3. 3Your browser opens the Hermes AI sign-in page. Sign in with Google or GitHub, then allow the request on the consent page.
  4. 4The client receives an access token issued for this MCP server, plus a refresh token so it stays signed in.

アプリができること

モデルを実行して前払いクレジットを使い、タスク・結果・残高を読み取ります。

アプリができないこと

API キー・請求・支払いの管理や、管理機能の利用。

いつでも接続解除

アカウントセキュリティに接続済みアプリがすべて表示されます。接続を解除するとすぐにアクセスが停止します。

接続済みアプリを管理

ツール

各ツールはコードと同じ API を呼び出し、料金・クレジットの確保・制限も同じです。

ツールDescriptionInputs
search_models
List Hermes AI API models for image, video, music and text generation. Filter by output type or a search query. Returns each model's slug, which create_task needs.
  • query string, optional
    Words to match in name, slug or description
  • output image | video | audio | text, optional
    Only models with this output type
get_model
Get one model's input fields (types, allowed values, defaults, limits) and its current price. Call this before create_task to build a valid input.
  • model string
    Model slug from search_models
create_taskSpends credits
Submit an asynchronous generation task. Charges the quoted price from the user's prepaid credits only if the task succeeds. Returns a task_id; then call wait_for_task or get_task. Reuse the same idempotency_key to retry safely after a network error.
  • model string
    Model slug
  • input object
    Fields from get_model
  • idempotency_key string, optional
    Optional; generated when omitted and returned for retries
  • expected_price_micros integer, optional
    Reject with PRICE_CHANGED if the quote differs
wait_for_task
Poll a task every 4 seconds until it finishes or the timeout passes, then return it like get_task. Call again if it is still running.
  • task_id string
    Task ID from create_task
  • timeout_seconds integer 5–120, optional
    Default 60
get_task
Get a task's status and results. Generated images are returned inline when ready.
  • task_id string
    Task ID from create_task
list_tasks
List the user's most recent tasks, newest first.
  • limit integer 1–100, optional
    Default 20
get_balance
Get the user's available, reserved and spent prepaid credits in USD.None

Troubleshooting

401 with “Sign in to use Hermes AI API”
The request carried no token, so sign-in has not finished. In Claude Code run /mcp and choose hermes-ai-api; in Codex run codex mcp login hermes-ai-api; in other clients choose Connect or Start next to the server.
401 with “The access token is invalid or expired”
The token expired, was revoked, or the app was disconnected in Account security. Sign in again from your client.
A tool returns INSUFFICIENT_CREDITS or another code
Tool errors carry the same status and code as the REST API. Add credits in Billing, or look the code up in the error reference.
wait_for_task returns while the task is still running
It waits at most 120 seconds (60 by default). Video and music can take longer; call it again with the same task_id.
get_model says the model cannot run
The model has no price configured right now. Pick another model from search_models; runnable shows which ones work.
My client has no OAuth sign-in
Send an API key instead: use the API key tab above, with the key in HERMES_API_KEY.

エラーコード請求Agent Skills