PlugWiki
ollaya logo

Ollaya MCP Server: Setup for Claude, Cursor and Other Apps

CommunityLocal
Getollaya-dev/ollaya
Tools
4
3 read · 1 change
Works in
3
supported apps
Risk
Medium
can write
Auth
None

Lets Claude, Cursor or another assistant ask small local models to classify, score or check a piece of text, without calling a large language model.

Ollaya is a program that runs open decision models locally, the way Ollama runs chat models. Its MCP server is one command of that program, so you install Ollaya first and then add the server to your assistant. It starts the local Ollaya service for you if it is not running.

The models answer typed questions with probabilities in milliseconds and never write text. Ollaya is made by Mert Cobanov and is not affiliated with Ollama or TypeSafe.

Good for

  • Developers who want fast routing, triage or moderation checks inside an agent
  • Teams who want decisions made on their own hardware, not a cloud API
  • Support and ops people sorting tickets or emails with an assistant

Works in

3 supported · 9 should work

Tap an app to see how to set it up and any known problems.

Install ollaya MCP

Claude Code

  1. 1

    Copy this prompt.

    Add the ollaya MCP server to Claude Code. Run this command:

claude mcp add ollaya -- ollaya mcp

Then check that it loaded and tell me if I need to sign in or restart.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires Ollaya installed (see README).

Install Ollaya first with the installer for your system from the README. The server is the ollaya mcp command and starts the local service if needed. Only the Claude Code command and the Cursor link come from the README; other configs are adapted from that command.

From the project README. Source.

Tools

  • Answers typed questions about a text or JSON

    Reads a message, email, ticket or any JSON and answers your questions about it with a local model. Each question is a choice, a score or a yes/no check, and you get probabilities back instead of generated text. You can pass your own questions or a named preset such as triage, email or moderation.

    Try: “Use the triage preset on this support email and tell me the intent and how urgent it is.”

    Read · decide

  • Lists the decision models you have installed

    Shows each model on your machine with its size and when it was downloaded. Use it to see which models the assistant can pick.

    Try: “Which Ollaya models do I have installed?”

    Read · list_models

  • Shows one model's details

    Returns a model's family, parameter count, context length, languages, capabilities, license and any built-in questions. It helps the assistant choose a model before it calls decide.

    Try: “Show me the license and languages of the laya model.”

    Read · show_model

  • Downloads a model from the Ollaya library

    Downloads a model by name, such as laya or decider, checks every file and reports progress. This is the only tool that writes to disk; models also download on first use by decide.

    Try: “Download the decider model so we can use it.”

    Write · pull_model

Read from the source code

Permissions

medium risk · 2 things to watch

Medium risk

Three tools only read, and pull_model downloads model files to your computer. It sends your text to a model on your own machine, not to a cloud service.

What it can do

  • Read your data
  • Create and change things
  • Delete things
  • Read-only mode

Watch out for

  • pull_model downloads files to your disk

    Models can be several gigabytes. Models also download on first use by decide, so ask the assistant to say which model it will use. github.com

  • It starts a local background service

    If nothing answers at the configured address, the server launches the Ollaya service on your machine. Set OLLAYA_HOST to a remote address to use one you run yourself instead. github.com

Security advisories

No published security advisories as of Nov 1, 2026.

Technical details
Talks toModel downloads when you call pull_model or when decide first needs a model (from the Ollaya library and the model author's Hugging Face repository)
Sign-in and keysNo credentials are used by the MCP server.
Scopes

  • None. No account or API key for the local setup.

Questions

Is the Ollaya MCP server official?

Yes, it ships with Ollaya itself as the ollaya mcp command. Ollaya is made by Mert Cobanov and is not affiliated with Ollama or TypeSafe.

Does it need an API key?

No. The models run on your machine. TYPESAFE_API_KEY in the README is only for the separate TypeSafe-compatible HTTP API, where any non-empty value works.

Does it send my text to the cloud?

No. The text goes to the Ollaya service on your computer. Only model downloads use the network.

Do I need a GPU?

No. The README says the multilingual laya model answers five questions in 516 ms on a CPU. The recommended winnow:e4b model is faster on an NVIDIA GPU.

Which assistants does it work with?

The README names Claude Code, Claude Desktop, Cursor and other MCP clients. It runs on your machine, so it does not work with hosted connectors such as ChatGPT.

How do I add it to Claude Code?

Install Ollaya first, then use the command in the install box above.

Information

updated Nov 2, 2026

Publisher
Community
Transport
stdio
Auth
None
License
Apache-2.0
Last commit
Oct 30, 2026
Open issues
2

WebsiteGitHubChangelogIssues

Sources

Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.

Show all 3
  1. README.md · github.com
  2. mcp.rs · github.com
  3. daemon.rs · github.com

Facts from the official MCP registry, GitHub, npm, PyPI and the Docker catalog. Updated Nov 2, 2026.