mcpField
H
HopBase

HopBase

Remote
hop-base.comAI & Agents0Updated 2d ago

About HopBase

HopBase is a hosted, read-only MCP server for coding agents using HopBase’s AI gateway. It reports key-specific model access, limits, account balance and usage without routing inference through MCP.

Key features

  1. 01List callable models and effective prices
  2. 02Inspect model endpoints and limits
  3. 03Search documentation by query
  4. 04View balance and spend-cap usage
  5. 05Review key details and usage metrics

Use cases

  1. 01Check which models an API key can call
  2. 02Look up a model’s documented request limits
  3. 03Review recent key spend and request totals

MCP support

Exposes read-only HopBase account and documentation tools to MCP clients over Streamable HTTP. Tool calls require the HopBase API key; model inference uses the regular API endpoints.

Configuration

Remote servers need no local install. Pick your client, then copy the config or install in one click.

Endpointhttps://api.hop-base.com/mcp
Copy install command
Claude CodeTerminal
Cursor.cursor/mcp.json
VS Code.vscode/mcp.json
Claude Desktopclaude_desktop_config.json
Windsurf~/.codeium/windsurf/mcp_config.json

Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.

All endpoints
Streamable HTTPhttps://api.hop-base.com/mcpAuth required

Tools

Details

From the project README

HopBase MCP server

HopBase is an AI gateway that serves Claude, GPT, Gemini, DeepSeek and other text, image and video models through one base URL. This MCP server lets a coding agent look up what its HopBase API key can do: which models it can call, their limits, its balance and its usage.

  • Endpoint: https://api.hop-base.com/mcp
  • Transport: Streamable HTTP, stateless, POST only (one JSON-RPC message per request; no SSE stream or batch requests)
  • Authentication: Authorization: Bearer <key> or x-api-key: <key>, with the same API key you use for model calls. Required to call a tool; listing the tools needs no key
  • Scope: read-only; every result covers only the key in use

This is a hosted service and its source is not published. This repository holds the MCP registry manifest (server.json) and client configs.

Tools

ToolReturns
list_modelsModels this key can call, with effective prices
get_model_limitsFor one model ID: whether this key can call it, plus the endpoints, request parameters, and limits documented for it, with links to the source pages
search_docsThe docs sections that best match a query (model ID, parameter name, error code), each with URL and snippet; lang selects en, zh-cn, zh-tw, es or ja
get_balanceAccount balance and this key's spend-cap usage, in USD
get_key_infoKey name, groups, spend cap, concurrency cap, expiry, and status
get_usageRequests, tokens, and billed cost for this key; last 7 days by default, up to 92 days

get_model_limits and search_docs read the published docs and refresh at most once an hour. The linked docs page is authoritative: a limit missing from a result does not mean there is none.

Setup

You need a HopBase API key: create one in the console and export it as HOPBASE_API_KEY.

Claude Code
claude mcp add --transport http hopbase https://api.hop-base.com/mcp \
  --header "Authorization: Bearer $HOPBASE_API_KEY"

The shell expands the variable when you run the command, so the key is saved in your local Claude Code configuration. For a shared project configuration, put examples/claude-code.mcp.json in the project as .mcp.json instead; Claude Code expands ${HOPBASE_API_KEY} at startup, so the file holds no key.

Cursor

Save examples/cursor.mcp.json as .cursor/mcp.json in the project, or as ~/.cursor/mcp.json for every project:

{
  "mcpServers": {
    "hopbase": {
      "url": "https://api.hop-base.com/mcp",
      "headers": { "Authorization": "Bearer ${env:HOPBASE_API_KEY}" }
    }
  }
}
Other clients

Any client that supports remote Streamable HTTP servers with a custom header can connect: point it at the endpoint and send the Authorization header above.

Good to know

  • The key decides what MCP sees. list_models and get_model_limits see only the groups of the key you configure. To inspect another model family, add its group to the key, or register a second server entry with a key that includes it.
  • Model calls do not go through MCP. Keep inference on the regular API endpoints (https://api.hop-base.com/v1: OpenAI Chat Completions and Responses, Anthropic Messages, native Gemini, and the image and video task endpoints).
  • Calling a tool needs a key. The handshake (initialize) and tools/list answer without credentials, so clients and directories can discover the tools; tools/call without a key returns 401.

Links

License

The files in this repository are released under the MIT License.

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.