Developer guide

Choose the right interface

• Use the catalog when a person needs to browse, filter, or compare models.
• Use Markdown when an agent needs compact page content with stable headings and links.
• Use the history API when software needs model metadata changes as JSON.
• Use the npm package when a Node.js application needs the catalog data locally.
• Do not use this catalog as the only source for a purchase or production routing decision. Provider prices, limits, and availability can change.

Markdown and discovery

Send Accept: text/markdown to a supported public page, or add .md to its path. Use /llms.txt for retrieval guidance and /sitemap.xml for the public page list.

curl -H 'Accept: text/markdown' https://llmcatalog.dev/models/openai/gpt-4o

History API

These read-only routes return JSON. The optional limit value must be from 1 through 1000.

GET /api/history/recent?limit=25
        GET /api/history/:provider/:model_id?limit=100

See the OpenAPI 3.1 document for request and response schemas.

Catalog lookup tools

POST /api/mcp accepts the documented tools/list and tools/call request shapes. Available tools are query_models, get_model, and list_providers. This is a small tool interface. It is not a complete MCP protocol server.

curl https://llmcatalog.dev/api/mcp \
        -H 'content-type: application/json' \
        -d '{"method":"tools/list"}'

JavaScript package

The official @agentjido/llmdb package provides the catalog for JavaScript and TypeScript applications. It requires Node.js 22.14 or later. The package does not currently install a command-line executable.

npm install @agentjido/llmdb

        import { llmdb } from "@agentjido/llmdb";
        const model = await llmdb.get("openai:gpt-5.4");

API errors

API failures use application/problem+json. Each response includes a stable code, HTTP status, human-readable detail, request instance, and a suggested resolution. Clients should branch on the code and status.

Data and service limits

Public interfaces have no authentication and no formal availability guarantee. Cache responses when practical. Check provider documentation before you depend on a price, feature, model name, or service limit.