Live model catalogfor coding assistants
MCP tools for Claude Code, Cursor, VS Code, and Codex to search, compare, cost-model, and live-test 300+ models.
Free to install. Live tests use your OpenRouter key.
Why this exists
Your assistant has no live model catalog
New models ship weekly. Pricing and capabilities change.
Without a catalog to query, your assistant falls back to training-time memory and web snippets. Pricing, limits, and model IDs in the answer can be wrong.
Index9 caches 300+ models, refreshes every 30 minutes, and exposes them as five MCP tools in your editor.
Workflow
How it works
Index9 adds 5 tools to your editor. Your assistant calls them when you ask about models.
You ask in chat
"Pick a model for the support agent in our backend: real quality on 4k-token threads at ~30k calls a day, without frontier per-token rates."
Your assistant calls index9
It searches live model data, compares finalists, and runs your prompt against the top picks.
find_modelsget_modelscompare_modelstest_modelYou get a measured pick
Backed by real cost numbers and real outputs, not training-data memory.
Real session
Live trace, not a mockup
Support-agent model pick: full tool trace and measured verdict.
“Pick a model for the support agent in our backend: real quality on 4k-token threads at ~30k calls a day, without frontier per-token rates.”
3.5K prompt · 500 completion · dry-run cost · 4 models tested
- 1find_models · sortBy=created, function_calling, minContext=8K, limit=30
- 2list_facets
- 3find_models · sortBy=relevance, function_calling + structured_output, "customer support quality instruction following", maxPrice=$5/M, minContext=32K, limit=20
- 4find_models · sortBy=price, function_calling + structured_output, maxPrice=$3/M, minContext=32K, limit=20
- 5compare_models · 4 finalists, 3.5K prompt + 500 completion
- 6test_model · dry-run × 4, 3.5K + 500
- $0.25/M in / $2/M out · $0.00162 per run
google/gemini-3.1-flash-lite2w agoTested - $0.04/M in / $0.18/M out · $0.00023 per run
openai/gpt-oss-120b10mo agoTested - $0.10/M in / $0.32/M out · $0.00051 per run
meta-llama/llama-3.3-70b-instruct1y agoTested - $0.11/M in / $0.22/M out · $0.00050 per run
deepseek/deepseek-v4-flash4w agoSelected
deepseek/deepseek-v4-flashA 284B MoE model (13B active) released April 2026 with reasoning, structured output, 1M-token context, and a per-call cost of $0.000504, just $15.12 per day at 30K calls, delivering frontier-class quality at a fraction of frontier per-token rates.
Tools
The 5 tools
Your assistant chains these together. You don't call them directly.
list_facetsLists available providers, capabilities, and modalities to filter by.
find_modelsSearches 300+ models by price, context size, capabilities, or natural language.
get_modelsPulls full specs and current pricing for any model.
compare_modelsDiffs 2–10 finalists side-by-side. Flags the cheapest for your token mix.
test_modelLiveRuns your prompt across models. Returns output, latency, cost. Dry-run for cost only.
Only test_model needs an OpenRouter key. The rest work out of the box.
Setup
Install Index9
Choose a client, copy the config, restart your assistant.
Cursor
{
"mcpServers": {
"index9": {
"command": "npx",
"args": ["-y", "@index9/mcp@latest"]
}
}
}Search, compare, and dry-run cost estimates work without a key. Live tests need an OpenRouter key.
FAQ
Common questions
07 entries