farebox-mcp
# farebox-mcp
**Farebox MCP Server** — connect Claude Desktop, Cursor, or any MCP-compatible client to 36+ frontier LLMs. Payments happen automatically per token in USDC on Solana — no accounts, no invoices.
[](https://www.npmjs.com/package/farebox-mcp)
[](./LICENSE)
## Install
```bash
npx farebox-mcp
```
No global install needed. Runs directly from the npm registry.
## Quick setup
### Claude Desktop
Edit `~/Library/Application Support/Claude/claude_desktop_config.json` (macOS)
or `%APPDATA%\Claude\claude_desktop_config.json` (Windows):
```json
{
"mcpServers": {
"farebox": {
"command": "npx",
"args": ["-y", "farebox-mcp"],
"env": {
"FAREBOX_API_KEY": "sk-fbx-your-key-here"
}
}
}
}
```
### Cursor
Edit `~/.cursor/mcp.json`:
```json
{
"mcpServers": {
"farebox": {
"command": "npx",
"args": ["-y", "farebox-mcp"],
"env": {
"FAREBOX_API_KEY": "sk-fbx-your-key-here"
}
}
}
}
```
Get your API key at [farebox.fun/dashboard](https://farebox.fun/dashboard).
## Keyless mode (x402)
No API key required — your agent pays per request autonomously in USDC on Solana:
```json
{
"mcpServers": {
"farebox": {
"command": "npx",
"args": ["-y", "farebox-mcp"],
"env": {
"FAREBOX_SOLANA_PRIVATE_KEY": "your-base58-private-key",
"FAREBOX_MODE": "x402"
}
}
}
}
```
## Available tools
Once connected, your agent has access to:
- **36+ frontier models** — GPT-4o, Claude Sonnet, Gemini Flash, Llama, DeepSeek, Mistral, and more
- **Community skills** — `summarize`, `translate`, `code-review`, `explain`, `sentiment`, `extract-data`, `draft-email`, `fix-grammar`
All models and skills appear as individual tools in your MCP client automatically.
## Environment variables
| Variable | Description |
|---|---|
| `FAREBOX_API_KEY` | API key from [farebox.fun/dashboard](https://farebox.fun/dashboard) |
| `FAREBOX_SOLANA_PRIVATE_KEY` | Base58 Solana private key (x402 keyless mode) |
| `FAREBOX_MODE` | `apikey` (default) or `x402` |
| `FAREBOX_BASE_URL` | Override API base URL (optional) |
## Pricing
Pay per token in USDC — no minimum, no subscription.
| Model | Input | Output |
|---|---|---|
| llama-4-scout | $0.11 / 1M tokens | $0.34 / 1M tokens |
| gpt-4o-mini | $0.15 / 1M tokens | $0.60 / 1M tokens |
| claude-sonnet-4-5 | $3.00 / 1M tokens | $15.00 / 1M tokens |
| gpt-4o | $2.50 / 1M tokens | $10.00 / 1M tokens |
Full model list: [farebox.fun/models](https://farebox.fun/models)
## Links
- Website: [farebox.fun](https://farebox.fun)
- MCP setup guide: [farebox.fun/mcp](https://farebox.fun/mcp)
- npm: [npmjs.com/package/farebox-mcp](https://www.npmjs.com/package/farebox-mcp)
- GitHub: [github.com/fareboxfun/farebox-mcp](https://github.com/fareboxfun/farebox-mcp)
- X: [@Farebox_](https://x.com/Farebox_)
## License
[MIT](./LICENSE)
TDQS
Scored across 5 tools
The tools are mostly distinct: list_models, chat, call_skill, get_balance, and get_usage each target a different concern (model discovery, inference, skills, billing, usage). However, get_balance and get_usage both deal with account-level billing/statistics and could be slightly ambiguous for an agent deciding which to call.
Tool names follow a consistent verb_noun pattern: list_models, call_skill, get_balance, get_usage. The only deviation is 'chat', which uses a bare noun-verb form rather than a verb_noun structure (e.g., send_chat or chat_completion), breaking pattern slightly.
Five tools is a well-scoped, focused set for an LLM API server. Each tool serves a clear purpose: discovery, inference, specialized skills, and account management. Every tool earns its place without redundancy.
The core workflow (discover models, chat, call skills) is well covered, and account management is present. Minor gaps exist: no explicit skill-list/discovery tool (skills are only implied via call_skill), and no streaming or multi-turn session management, but these are reasonable to omit for a lightweight connector.