AI Model Research

A Claude Code plugin for researching, filtering, comparing, and evaluating AI models on OpenRouter. Pulls live data from the OpenRouter catalog and enriches it with external research where appropriate.

This plugin subsumes two earlier projects:

• The standalone open-router-model-research plugin (8 catalog-fetching skills).
• The Model-Scout-MCP server, which is now bundled as the plugin's MCP backbone — providing in-process catalog caching and two consolidated tools (get_model, consider_models).

What's inside

MCP server

Server	Provides
model-scout	get_model (direct ID lookup) and consider_models (filter / compare / recommend / cost / alternatives). Caches the OpenRouter catalog for ~10 minutes.

Skills

Skill	What it does
or-lookup-models	Foundation skill — fetches the OpenRouter catalog or a single model's record.
or-quick-lookup	Lightweight mid-conversation snapshot of a single model (capability + pricing) via the MCP.
or-find-tool-models	Filters the catalog to models supporting tool use / function calling.
or-find-vision-models	Filters to vision-capable (image-input) models.
or-find-audio-models	Filters to models that accept audio input.
or-recommend-model	Interactive recommendation — asks about budget, context, modalities, then proposes a ranked shortlist.
or-compare-models	Builds a head-to-head side-by-side comparison of 2+ models.
or-evaluate-model	Deep evaluation of a single model — combines catalog data with model card, paper, license, benchmarks.
or-research-models	Open-ended exploration of the catalog around a theme, family, or use case.
or-cost-projection	Estimate the cost of running a workload (requests × tokens) on one or more models.
or-find-alternative	Find cheaper / similar alternatives to a reference model with explicit trade-off framing.

Why this plugin

The OpenRouter catalog is large and changes weekly. These skills:

• Always pull live catalog data — no stale recommendations from training memory.
• Share an in-process catalog cache via the bundled MCP, so repeated lookups don't re-hit the API.
• Cover the whole research lifecycle: discover → filter → recommend → compare → evaluate → cost → alternatives.
• Look beyond the OpenRouter catalog to underlying model cards, papers, and licenses where appropriate.

Installation

claude plugins install ai-model-research@danielrosehill

No API key required — the OpenRouter catalog endpoint (/api/v1/models) is unauthenticated, and so are all the skills and MCP tools in this plugin. (You only need an OPENROUTER_API_KEY if you want to actually call a model through OpenRouter, which this plugin doesn't do — it's research-only.)

Usage

The skills fire automatically when relevant. Example prompts:

• "Recommend an OpenRouter model for legal document summarization with a budget of $1/M tokens"
• "Compare claude-sonnet-4-5 and gpt-4o on OpenRouter"
• "What vision models are available on OpenRouter under $0.50/M?"
• "How much would 50k requests at 2k in / 500 out tokens cost on claude-sonnet-4.5?"
• "What's a cheaper alternative to gpt-4o with similar tool-use quality?"

License

MIT