Models and AI infrastructure

All 41

Alphabetical. Checked on 2026-09-04; anything that did not answer says so.

• AI Dev Jobs MCP the site refused an automated request
  The same job board exposed over MCP so an agent can search it. The endpoint refused a plain request.
  A job board with a tool interface.
• Clarifai the connection timed out
  Computer vision and model deployment APIs. The site timed out on every attempt.
  Unreachable: the connection timed out.
• Claude (Anthropic)
  Anthropic's model family, and the one this catalogue's own tooling is written against.
  The model behind a good deal of this list.
• Claude MCP Servers the domain does not resolve
  A directory of MCP servers. The domain no longer resolves.
  Gone: the domain does not resolve.
• CLIRank
  Scores APIs on how usable they are without a human: auth method, headless operation, pricing transparency.
  Rates APIs for the machine, not the reader.
• CloudSight
  Image recognition returning natural language descriptions rather than label sets.
  Answers in sentences, not in tags.
• Code Llama
  Meta's open-weight code model family, released with inference code and licence terms permitting commercial use.
  Made a serious code model something you could download.
• Codestral
  Mistral's code model, available through their platform and for local deployment under their licence.
  Europe's entry in the code model race.
• Cursor MCP Directory
  A community index of MCP servers and plugins usable from Cursor.
  The plugin list, before there was a plugin list.
• DeepSeek Coder
  An open-weight code model that closed much of the gap to proprietary systems, released with weights and paper.
  The release that made 'open weights are behind' harder to say.
• DialogFlow
  Google's conversational agent platform, now running on Gemini, with the older intent model still underneath.
  Survived the shift from intents to models.
• Gemini (Google)
  Google's model family, with very long context and native multimodal input.
  Competes on how much you can put in.
• Google Vision AI
  Pre-trained image and video recognition APIs: labels, text, faces and objects without training a model.
  Vision without a training run.
• GPT-5 (OpenAI)
  OpenAI's frontier model line. The catalogue entry points at the company rather than at a product page.
  The entry is a company, not a tool.
• GPT4All
  Runs open models locally on ordinary hardware, with a desktop application rather than a command line.
  Local models for people who will not compile anything.
• Haystack
  Modular building blocks for retrieval and agent pipelines, aimed at production rather than at demonstrations.
  Components, not a framework that owns your program.
• Helium MCP the page returns 404
  An MCP server for news with bias scoring. The repository returns 404.
  Gone: the repository is not there.
• Jan
  An open-source desktop client that runs local models or connects to hosted ones, with the same interface for both.
  Local and cloud behind one window.
• LangChain
  The framework that popularised chaining model calls with tools and memory, now positioned as an agent engineering platform.
  Defined the vocabulary, then had to defend it.
• LiteLLM
  A gateway that speaks the OpenAI format to a hundred providers, with cost tracking, guardrails and load balancing.
  Makes provider choice a configuration line.
• llamafile
  Packages a model and its runtime into a single executable file that runs on several operating systems without installation.
  A model you can email.
• LlamaIndex
  Data framework for connecting documents to models, including the parsing of formats that resist extraction.
  Started at ingestion, which is where retrieval actually fails.
• LM Studio
  A desktop application for running local models, with a server mode that speaks the OpenAI API.
  Local models behind the API everything already speaks.
• LocalAI
  An open-source drop-in replacement for the OpenAI API that runs models locally, including without a GPU.
  The same API, on hardware you already own.
• MCP Market
  A directory of MCP servers connecting assistants to tools like Figma, Databricks and Ghidra.
  The catalogue that appeared as soon as the protocol did.
• MCP Server Finder
  Another index of MCP servers, searchable by capability.
  One of several directories competing to be the registry.
• MCP.so
  An MCP server directory with a focus on connecting applications to tools and workflows.
  Short domain, long list.
• MCPServers.Net
  Listed as an MCP directory. The domain is now parked and offered for sale.
  Gone: the domain is for sale.
• Not Human Search
  A search engine indexed for agents rather than people, covering tools, MCP servers and APIs.
  Search built for the reader who is not a reader.
• Ollama
  Runs open models locally with one command, and serves them over a simple API. The reason most people's first local model worked.
  Made running a model as easy as pulling an image.
• Omni Skills Forge the page returns 404
  A skill manager for assistants. The repository returns 404.
  Gone: the repository is not there.
• OpenRouter
  One API across nearly every model, with prices and availability compared in the open.
  Turned the model market into a price list.
• Phi-4 (Microsoft)
  A fourteen-billion parameter Microsoft Research model that performs above its size on reasoning, largely through training data quality.
  An argument that data beats parameter count.
• Promptfoo
  Testing, evaluation and red-teaming for prompts and agents, including vulnerability scanning.
  Treats a prompt as something you can regression-test.
• PulseMCP
  A daily-updated directory of MCP servers, one of the more actively maintained indexes.
  Updated daily, which most directories are not.
• Qwen2.5-Coder
  Alibaba's open-weight code model line. The repository has moved on to Qwen3-Coder.
  Superseded by its own next version.
• Roboflow
  The full computer vision pipeline: annotation, training, evaluation and deployment.
  Covers the labelling, which is where vision projects die.
• Semantic Kernel
  Microsoft's SDK for embedding models into applications, with planners and plugins as first-class parts.
  Enterprise-shaped from the beginning.
• StarCoder 2
  An open code model from the BigCode project, trained on permissively licensed source with the data provenance published.
  Published what it was trained on, which almost nobody does.
• Vercel AI SDK
  A duplicate entry for the AI SDK, filed under models rather than assistants. One toolkit, two rows.
  The catalogue lists it twice.
• VS Code MCP Directory
  The GitHub MCP Registry, centralising MCP servers for discovery from VS Code and elsewhere.
  The registry attempt from the platform that hosts the code.

Other shelves