recommend_modelLABSCO SUMMARY
Model pricing that is current, answered as a routing decision.
A model's own knowledge of LLM prices is stale the day it ships, which makes cost reasoning inside an agent unreliable. This answers with data refreshed every four hours, and takes the shape of the actual decision: a task, a budget, a set of capability requirements, and a pick with its reasoning.
What it is
A model routing advisor served over MCP by whichmodel.dev, which tracks pricing and capabilities across 100+ LLM models and refreshes them every four hours.
What you get
- A cost-optimised recommendation for a task type — chat, code generation, code review, summarisation, translation, data extraction, tool calling, creative writing, research, classification, embedding, vision or reasoning
- Complexity, expected input and output tokens, and a per-call budget taken as inputs, with capability requirements like tool calling, JSON output, streaming, minimum context window and provider include/exclude lists
- A recommendation, an alternative, a budget option, a cost estimate and the reasoning behind the pick
- Head-to-head comparison of two to five models by ID, with pricing, capabilities, quality tiers and projected daily and monthly costs at your call volume
- Raw pricing lookup filtered by model, provider, price ceiling and capabilities
- Price changes checked, for catching a provider's move before it lands on the bill
Requirements
Nothing — no key, no install. Point your client at https://mcp.whichmodel.dev/mcp over streamable HTTP, or run npx -y whichmodel-mcp as a local stdio proxy with WHICHMODEL_URL set to that address. The repository's server.json lists that address; the README still gives https://whichmodel.dev/mcp, which no longer answers, and the proxy falls back to it unless WHICHMODEL_URL says otherwise.
Setup effort
Add the hosted endpoint — paste it into your client
USE CASES
01
Choosing a model for a job you are about to run
Give the task type, complexity, expected token counts and per-call budget and get a pick, an alternative and a budget option with a cost estimate.
02
Deciding between two models you are already considering
Compare them head to head on pricing, capabilities and quality tier, with cost projected at your daily call volume.
03
Working out what a feature will cost at scale
Project daily and monthly cost from calls per day and average input and output tokens before you commit to a model.
04
Filtering to models that meet a hard requirement
Constrain by tool calling, JSON output, streaming, minimum context window or an allowed provider list, so the recommendation is one you can actually use.
05
Catching a price change before the invoice does
Check what has moved across the tracked models rather than finding out from billing.
4 tools
Listed in project docs. Your client may show a different list.
compare_modelsget_pricingcheck_price_changesBEFORE YOU INSTALL
What to know first
Tools you get
4 tools named in the project's documentation.
Requires
npx on PATH (Node.js 20 or newer) for the stdio proxy in the client configs — or a client that can reach https://mcp.whichmodel.dev/mcp over streamable HTTP directly
Where it runs
Both documented clients run the server on your machine.
INSTALLATION
Connect your client
Server setup
ENDPOINT URL
https://mcp.whichmodel.dev/mcpAdd this hosted endpoint to your MCP client.
MCP Clients
Claude Desktop, Cursor
JSON
{
"mcpServers": {
"whichmodel": {
"command": "npx",
"args": [
"-y",
"whichmodel-mcp"
],
"env": {
"WHICHMODEL_URL": "https://mcp.whichmodel.dev/mcp"
}
}
}
}
