v1.1.0 — package rename, compare() fixes, richer cost estimator

What's new in v1.1.0

Packaging

• Renamed the Python package server/ → quelllm_mcp/ with a proper main() console entry point. The old top-level server package name risked import collisions on user machines. Config and install are unchanged (command: "quelllm-mcp").

Bug fixes

• compare(): removed the tokSec field (it never existed in the quelllm.fr API, so it was always null) and exposed the real ctx (context window) instead. Replaced the bogus more_recent placeholder — which actually compared parameter counts — with honest more_params and larger_context signals.
• _fetch_json: now serves a stale cache entry when quelllm.fr is unreachable, instead of crashing the tool call with a raw stack trace.

Cost estimator

• estimate_cost gained prompt-caching, batch mode, DeepSeek off-peak discounts, a token-capacity check for self-hosted hardware, recoverable-VAT and pc_always_on toggles — matching the web calculator at https://quelllm.fr/calculateur-cout-llm.

Docs / metadata

• Removed the phantom PyPI package declaration from server.json (the package is not published on PyPI). Documented install via pip install git+… and uvx.
• Corrected the catalog size claim from "250+" to "190+" across README, pyproject.toml and server.json.

Install

pip install git+https://github.com/MGM-FALCON/quelllm-mcp.git

v1.0.0 — Initial release

First public release of quelllm-mcp.

Tools

• list_models(filter_origin?, filter_family?, max_params_b?) — list open-weights LLMs from quelllm.fr (250+ models)
• get_model(model_id) — full record (params, vram, tokSec, license, install)
• compare(a, b) — side-by-side comparison
• estimate_vram(model_id, quant) — VRAM in GB + recommended GPU/Mac tiers
• estimate_cost(input_tokens_per_month, output_tokens_per_month, ...) — full table API providers vs self-hosted (in EUR)
• search_models(query) — fuzzy search

Install

pip install git+https://github.com/MGM-FALCON/quelllm-mcp.git

Use with Claude Code / Desktop / Cursor / Continue

{ "mcpServers": { "quelllm": { "command": "quelllm-mcp" } } }

Data

All data pulled from https://quelllm.fr/api/ (CC BY 4.0, no key, CORS). Cached locally 1h.