Browse a model gateway's catalog and per-token prices from chat, then run prompts on any listed model.
Picking a model for a job is mostly a pricing and limits question. Which one is cheapest per million input tokens? How large is its context window? Does it support the parameter you need? The UplinkSwitch MCP server lets an assistant answer those questions directly from the gateway's live catalog, before anyone has signed up for anything.
UplinkSwitch is a pay-per-token gateway. One key and one base URL reach Claude, GPT, Gemini, Grok, DeepSeek, GLM, Qwen, Kimi and various open-weight models, using the OpenAI, Anthropic and Google wire formats. Billing runs on prepaid credit with no subscription or monthly minimum.
| Tool | Key needed | Use |
|---|---|---|
list_models |
no | every model with its price per million tokens, filterable by text or kind (chat, embedding, rerank, transcription) |
model |
no | one model's context window, output limit, supported parameters, endpoints and exact pricing |
compare_prices |
no | prices sorted cheapest input first, optionally filtered by a tag such as reasoning |
chat |
yes | runs a conversation on a chosen model and returns the answer, token usage and request cost |
That split is the practical point. Three of the four tools are read-only catalog lookups and work anonymously. Only chat spends money.
The server is published on npm as @uplinkswitch/mcp and runs with npx -y @uplinkswitch/mcp, so Node.js is the only prerequisite. Building from source with npm install && npm run build also works.
Two environment variables control it:
UPLINKSWITCH_API_KEY is required only for chat. Leave it out if you just want price research.UPLINKSWITCH_BASE_URL swaps the upstream host. It defaults to https://api.uplinkswitch.com/v1, and the same network runs several sibling brands on their own hosts. The catalog tools work against any of them without a key, which makes cross-brand price comparison possible.It fits developers who are evaluating models for a project and want the assistant to do the price legwork, and teams already on the gateway who want to test prompts against different models without leaving their editor.
Treat the figures as indicative. Gateway prices change, so confirm current rates against your own account before committing a budget. Every chat call is billed to your credit, and a curious agent looping through models can burn through it faster than you expect.
If you also want your assistant to read up-to-date library documentation, Context7 pairs naturally with a model gateway. More utilities like this live in our developer tools section.
Check the linked publisher documentation and your connected client's tool list. Names and availability can vary with the server version and account permissions.
| Tool | What it does |
|---|---|
| list_models | List every model with its price per million tokens, filtered by text or by kind (chat, embedding, rerank, transcription). |
| model | Return one model's full entry: context window, output limit, supported parameters, endpoints and exact pricing. |
| compare_prices | Show prices per million tokens sorted cheapest input first, optionally filtered by tag. |
| chat | Run a conversation on a chosen chat model and return the answer, token usage and request cost. Requires an API key. |
{
"mcpServers": {
"uplinkswitch": {
"command": "npx",
"args": [
"-y",
"@uplinkswitch/mcp"
],
"env": {
"UPLINKSWITCH_API_KEY": "sk-uplink-live-..."
}
}
}
}Add to claude_desktop_config.json. The API key is only needed for the chat tool; remove the env block for catalog-only use.
| Variable | Description | Required |
|---|---|---|
| UPLINKSWITCH_API_KEY | API key for the gateway. Only the chat tool needs it; the catalog tools work without one. | Optional |
| UPLINKSWITCH_BASE_URL | Upstream API host. Defaults to https://api.uplinkswitch.com/v1; change it to query a sibling host on the same network. | Optional |
Read, search and edit files inside allowed directories with the Model Context Protocol reference Filesystem server.
Retrieve version-aware library documentation and code examples with Context7 MCP, using resolve-library-id and query-docs.
AI-native project management with persistent memory across IDEs (Claude, Cursor, Codex).
⭐ sifa-mcp — Portable Reputation & Skills Passport MCP Server
PreClick is a URL security scanner and MCP server that enables AI agents and any client to analyze URLs for phishing, malware, and other security
A Model Context Protocol (MCP) server designed to interface with SIEMENS WinCC Unified SCADA systems via their GraphQL API. This server exposes