Woolly MCP server
Commit Nov 15, 2026Provides a shared, local memory for AI applications using plain Markdown files to store and recall personal facts and project notes.
Also known as rjcuff/woolly
- Stars
- 12
- Tools
- 8
- Downloads, 30 days
- —
- Latest version
- 0.13.1
- Last commit
- Nov 15, 2026
- License
- MIT
What is Woolly MCP?
Woolly provides a shared, local memory for AI applications using plain Markdown files to store and recall personal facts and project notes. It exposes 8 tools, such as remember, recall, decide.
Summary generated from the project's README
About Woolly MCP server
Woolly provides a shared, local memory system for various AI applications by storing information in plain Markdown files. It allows different tools like Claude Desktop, Cursor, VS Code, and Windsurf to access a unified set of personal facts and project notes. Because the data stays on your machine as local files, your information remains private and does not leave your computer.
Developers can integrate this memory into their workflows using the woolly-ai Python package or by connecting it to MCP-compatible clients. The system works with Ollama to enable semantic search and can be configured as a proxy for apps using OpenAI or Anthropic APIs. It manages different contexts by keeping project-specific notes separate from general personal facts, ensuring that an AI working on one codebase does not access notes from another.
Users can manage their stored data through a built-in UI or via the command line. Commands like woolly learn use local models to extract facts from conversations, while woolly sync allows for encrypted data synchronization across devices on a local network without using a cloud service.
What you can do with it
- 1Unify memory across different AI toolsConnect multiple AI applications so they all share the same personal context and project information.
- 2Automate fact extraction from chatsUse a local model to identify and save important details from your conversations automatically.
- 3Synchronize memory across multiple devicesKeep your personal AI context consistent between a laptop, desktop, and other hardware over your own network.
- 4Import existing AI chat historiesBring your previous knowledge from other platforms into your local memory system.
Written from the project's README. Every use case is checked against a quote from it.
Woolly MCP tools 8
Read from the server with tools/listrememberSave a durable fact to Woolly, the user's local memory. Personal memory is about the user (preferences, how they work, facts about them) and stays on their computer. Project memory is about the codebase: its stack, conventions, decisions, gotchas and commands. In a repo with .woolly/woolly.md, project memory is shared with the whole team through git, so only save what every teammate's agent should know. Give a stable key for facts that can change; a newer memory with the same key replaces the older one.
recallSearch Woolly for anything relevant to a query: the user's personal memory plus the current project's memory, which may include notes from teammates. Returns current facts, best match first.
decideSave a decision with the reason and the options that were turned down, so no app suggests them again. Use when the user chooses an approach or rejects one.
handoffSave a short note on where this session left off: what you were working on, what got done, what was decided and what's next. The next session in any app starts with it. One per project; a new one replaces the last.
get_profileGet a short summary of everything currently known about the user: facts, preferences and notes. Good to call once at the start of a conversation.
list_memoriesList the user's memories, newest first.
forgetMove a memory to the trash by id. Only do this when the user asks. The user can restore it.
resolve_conflictMake one memory the current fact for its key, replacing the others. Only do this when the user confirms which is right.
| remember | Save a durable fact to Woolly, the user's local memory. Personal memory is about the user (preferences, how they work, facts about them) and stays on their computer. Project memory is about the codebase: its stack, conventions, decisions, gotchas and commands. In a repo with .woolly/woolly.md, project memory is shared with the whole team through git, so only save what every teammate's agent should know. Give a stable key for facts that can change; a newer memory with the same key replaces the older one. | |
| recall | Search Woolly for anything relevant to a query: the user's personal memory plus the current project's memory, which may include notes from teammates. Returns current facts, best match first. | |
| decide | Save a decision with the reason and the options that were turned down, so no app suggests them again. Use when the user chooses an approach or rejects one. | |
| handoff | Save a short note on where this session left off: what you were working on, what got done, what was decided and what's next. The next session in any app starts with it. One per project; a new one replaces the last. | |
| get_profile | Get a short summary of everything currently known about the user: facts, preferences and notes. Good to call once at the start of a conversation. | |
| list_memories | List the user's memories, newest first. | |
| forget | Move a memory to the trash by id. Only do this when the user asks. The user can restore it. | |
| resolve_conflict | Make one memory the current fact for its key, replacing the others. Only do this when the user confirms which is right. |
Configuration
OLLAMA_HOST
| Name | Description | |
|---|---|---|
| OLLAMA_HOST | — |
Automated checks
Ran Nov 13, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachablePass
- Open source license (OSI)Pass
- Commit or release in the last yearPass
- Package name not close to a popular packageNot checked
- Remote endpoint uses HTTPSNot checked
- Destructive tools are labelledNot checked
- No hidden instructions to the modelPass
- No critical advisory in the package or its dependenciesNot checked
README
View on GitHubWoolly
One memory for all your AI tools. Local, private, and always up to date.
Ollama runs your models. Woolly remembers for them.
Website · Docs · Quickstart · Bluesky · LinkedIn

Every AI app forgets you, or keeps its own memory on someone else's servers. Claude doesn't know what you told ChatGPT, and Cursor doesn't know either.
Woolly runs on your computer and gives all of them one shared memory. It's stored as plain Markdown files you can open and edit, and nothing ever leaves your machine.
Get started
curl -fsSL https://usewoolly.com/install.sh | sh
woolly remember "I prefer short answers"
woolly connect
woolly connect adds Woolly to Claude Code, Claude Desktop, Cursor, VS Code, Windsurf, Codex, Gemini CLI, LM Studio and opencode. Now they all remember.
Using Claude Desktop? Download woolly.mcpb and double-click it.
Other ways to install: brew install rjcuff/tap/woolly, or on Windows irm https://usewoolly.com/install.ps1 | iex, or go install github.com/rjcuff/woolly/cmd/woolly@latest.
What it does
Works with your AI tools. Claude Code, Claude Desktop, Cursor and any other MCP app can save and look up memories on their own.
Gives Ollama a memory. woolly proxy adds what Woolly knows to every chat, for Ollama and any app that uses the OpenAI or Anthropic API.
Keeps facts current. Say you moved to Denver, and Denver replaces Austin. Your AI stops bringing up things that stopped being true.
Lets you decide. If an AI guesses something that contradicts what you said, your version stays and the guess is flagged for you to review.
Loads by itself. In Claude Code, every session starts with your profile and notes about the project you're in. No need to ask.
Learns from your chats. woolly learn uses a model running in Ollama to pick out facts from a conversation. Newer facts replace older ones, and nothing leaves your computer.
Syncs your devices. woolly sync keeps your laptop, desktop and Raspberry Pi in step over your own network. Encrypted, no cloud, no account.
Keeps projects apart. Notes about a codebase stay with that repo, and facts about you go everywhere. Your AI in one project never gets another project's notes.
Shows you everything. woolly ui opens a page where you can search, edit and delete every memory. Or open the files yourself. Every memory notes which app saved it, and woolly log shows what each app saved, read and deleted.
Brings in your history. woolly import pulls facts about you out of your ChatGPT and Claude exports.
Stays private and safe. No cloud, no account, no tracking. Passwords and API keys are refused, an AI can't plant instructions in your memory, and deleted memories go to a trash you can restore from.
Builds into your apps. pip install woolly-ai and add memory to any prompt in one line.
Give Ollama a memory
woolly service install # or run `woolly proxy` in a terminal
OLLAMA_HOST=127.0.0.1:11436 ollama run llama3.2
Every chat now includes what Woolly knows about you, and facts you mention are saved. For OpenAI or Anthropic apps, point the app's base URL at the proxy. See Proxy.
Use it from Python
pip install woolly-ai
import woolly
woolly.remember("Moved to Denver in October", key="user.location")
system_prompt = "You are a helpful assistant.\n\n" + woolly.context(user_message)
See Python.
Commands
woolly remember "Moved to Denver" --key user.location # a newer fact with the same key replaces the old one
woolly recall "where do I live"
woolly profile # a summary of what Woolly knows about you
woolly ui # review and edit memories in your browser
woolly log # what each app saved, read and deleted
woolly learn "I just moved to Denver" # a local model picks out the facts
woolly sync init # keep your devices in step
woolly import chatgpt ~/Downloads/export.zip --dry-run
woolly forget <id> # moves it to the trash; undo with woolly restore <id>
woolly proxy # memory for Ollama, OpenAI and Anthropic apps
woolly connect # add Woolly to every AI app on this computer
woolly service install # run in the background and start at login, like Ollama
Full list: Commands.
Connect your apps
woolly connect
Or add this to any MCP app's config yourself:
{ "mcpServers": { "woolly": { "command": "woolly", "args": ["mcp"] } } }
See Connect your apps.
Do I need Ollama?
No. Woolly works on its own. If Ollama is running, Woolly also uses it to search by meaning.
Repo layout
| Folder | Contents |
|---|---|
cmd/woolly, internal |
The Woolly program, written in Go |
docs |
Documentation in MDX |
ui |
The review page, built into the binary |
sdk/python |
The woolly-ai Python package |
web |
Website and docs site |
scripts |
Install, release and Homebrew scripts |
packaging/homebrew |
The Homebrew formula, kept in step with the tap |
packaging/mcpb |
The Claude Desktop bundle |
More
MIT license.
Woolly MCP questions
What tools does Woolly MCP provide?
It exposes 8 tools, including remember, recall, decide, handoff, get_profile.
Does Woolly require Ollama to work?
No, Woolly works on its own, but if Ollama is running, Woolly uses it to perform semantic searches by meaning.
How is my data stored and kept private?
Data is stored locally as plain Markdown files on your computer, ensuring nothing leaves your machine.
Can I sync my memories across different devices?
Yes, you can use the sync command to keep your devices in step over your own network without using a cloud service.
How can I use Woolly with my existing AI tools?
You can use the connect command to add Woolly to apps like Claude Desktop, Cursor, and VS Code, or use the proxy for Ollama and OpenAI/Anthropic APIs.