Ask Claude "how much does this model cost?"
Install once. Get live LLM pricing tools inside Claude Desktop, Claude.ai custom connectors, Cursor, Windsurf, Zed, Cline, and any other MCP client. Free. Read-only. No auth.
What it does
A read-only MCP server backed by our live pricing table. Six tools, no surprises.
-
list_modelsList LLM models with prices, filter by provider, category, or verified-only.
-
get_modelFull record for one model: input/output/cached prices, context window, last-verified date.
-
compare_modelsSide-by-side price comparison for 2-6 models.
-
estimate_costDollar cost for a given input/output token count on any model. Handles cached-input pricing.
-
cheapest_forTop-N cheapest verified models in a category (fast / flagship / reasoning / embedding).
-
list_providersCatalogue of every provider we track with their official pricing-page link.
Two resources are also exposed:
usagewall://pricing/full (the entire table) and
usagewall://pricing/<provider> (one provider, e.g.
usagewall://pricing/openai).
Install in Claude
Open Claude settings
Claude Desktop → Settings → Developer → Edit Config.
Claude.ai → Settings → Connectors → Add custom connector.
Add the server
For Claude.ai custom connectors paste the URL below. For Claude Desktop config, see the JSON snippet.
Ask a question
"What's the cheapest reasoning model right now?" or "Cost of 100K input + 5K output on Claude Sonnet 4.6?"
Claude.ai custom connector
Paste this URL into the custom connector dialog:
https://usagewall.vercel.app/api/mcp
Claude Desktop config
Edit ~/Library/Application Support/Claude/claude_desktop_config.json
(macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):
{
"mcpServers": {
"usagewall": {
"url": "https://usagewall.vercel.app/api/mcp"
}
}
}
Cursor / Windsurf / Zed / Cline
Same URL, drop it in the MCP server settings of any of these clients. UsageWall speaks the standard HTTP MCP transport — no SDK required, no auth handshake.
Quick check from the terminal
List tools without any client:
curl -s https://usagewall.vercel.app/api/mcp \ -H "Content-Type: application/json" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
Estimate cost of 50K input + 2K output on GPT-4o:
curl -s https://usagewall.vercel.app/api/mcp \ -H "Content-Type: application/json" \ -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"estimate_cost","arguments":{"provider":"openai","model_id":"gpt-4o","input_tokens":50000,"output_tokens":2000}}}'
What's not here
- No write endpoints. Nothing in this server changes any state.
- No auth, because there's nothing private to read. We're considering OAuth for a future Plus tier that exposes your own spend data.
- No streaming responses for now. Tool calls return their full result in one shot — fine for pricing-sized payloads.
Source & reliability
Built on our public pricing JSON, hand-checked
monthly against the providers' own pricing pages plus an
OpenRouter
live overlay. Models pending official pricing are marked
verified: false. Bugs and price corrections welcome at
hello@agenciabaku.com or on
GitHub.