The Curator MCP
A local wiki memory built from what you read: a shared knowledge folder for agents through an MCP bridge and my-curator.
- 103 ★ on the repository
- Last commit
Extended access
Why this level
- The bridge reads and writes knowledge and project state files and source text goes to the model provider for wiki queries.
Install
- Works in
Manual install
curl -fsSL https://raw.githubusercontent.com/talirezun/the-curator/main/install.sh | bashA Mac option with no download. There is a .dmg app, one line for Windows PowerShell and a manual setup for Linux.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The bridge reads and writes knowledge and project state files and source text goes to the model provider for wiki queries.
This is third-party code. Review the repository files before installing.
What it does
The Curator turns articles, notes and transcripts into a connected personal wiki: a page for every person, tool and idea. Each new source updates existing pages instead of creating a copy, and you ask the wiki questions in ordinary language and get answers pointing to the pages. Everything lives in a folder on your computer, with no account and no one else's server. The same folder can be built together with a team through the optional Shared Brain. For agents there is a working state layer, called Memory in the app: an agent writes where it stopped at the end of a session and at the start of the next one receives the state and the project's canonical documents in a single MCP call. The bridge runs as a stdio child process for Claude Desktop, Claude Code and Cursor. The `my-curator` command with `context`, `save`, `doctor` and `install-hooks` gives the same from a shell with the app closed. Two skills and hooks come with it for harnesses where they fire.
Who it is for. People who read a lot and run projects with agents: they need a compounding wiki and state that survives a change of session, model and machine.
Good fit when
- Work runs for weeks and a new agent session must know where you stopped and what was already tried.
- Articles and transcripts should become a wiki that updates with each source.
- A project's architecture and decisions should travel with it across machines.
- A team keeps a shared knowledge folder through a private GitHub repository.
Not a fit when
- The agent runs in a browser: the bridge launches as a local program that a browser-only assistant cannot reach.
- You need a guaranteed save: state capture stays advisory and a missed save returns the previous state.
- You need plain MIT licensing for all code: part of the files use a source-available license with restrictions.
Example request
Load the project state and canonical documents through Curator, continue from where we stopped and save a new handoff at the end of the session.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
The MCP bridge launches as a child process on your computer, so a client must be able to start a local program. State capture is advisory: nothing forces an agent to save and a missed save returns the previous state, stale but not corrupted. Skill activation depends on the harness: in one measured harness the skill alone saved 0 of 4 times and hooks lifted that to 3 of 4. Mac has a .dmg app without notarization, Windows installs through a PowerShell command and Linux runs as a local server in the browser. Node.js 18 is required and wiki work needs a model API key. The license combines MIT with source-available terms.
How to disable. Remove the bridge entry from the client config and quit the app. `my-curator doctor` shows what is wired on this machine for hooks. The knowledge folder stays as plain files.
FAQ
How do I install The Curator?
On Mac download the .dmg app or run `curl -fsSL https://raw.githubusercontent.com/talirezun/the-curator/main/install.sh | bash`. On Windows paste one line into PowerShell and on Linux run the local server.
How do I connect an agent?
In Settings open MCP bridge, then Agent apps and press Install for me. The bridge runs as a child process and does not need the web app.
What does my-curator do?
`context` prints a project's bootstrap, `save` writes a complete handoff from standard input, `doctor` reports what is wired and `install-hooks` wires a harness.
How does working state differ from the wiki?
The wiki accumulates, state is replaced whole at each save so a resolved blocker does not return and a canonical document is kept verbatim.
Is team sync required?
No, Shared Brain is optional. Without it everything stays in a folder on your computer and sync goes through your private GitHub repository.
Discussion
No questions or reviews yet.
Related
MCP serversloci turns notes and agent memories into one knowledge base with hybrid search and answers cited down to a file section.
Third Brain Skills: a knowledge base and verification for agents
Third Brain Skills
21 skills build an Obsidian-style knowledge base with provenance, daily loops and verification before any done claim
Persistent agent memory in an Obsidian vault: saving decisions, search, research and background vault upkeep
An agent saves web pages, PDFs and scans into a local vault, builds linked concepts and answers from them with source references.
Four skills for a personal Obsidian knowledge base: an agent compiles articles and notes into a linked wiki, answers from it and health-checks it
An Obsidian vault template with 18 skills, 72 commands and 6 subagents for Claude Code: the agent turns sources into linked pages and answers from them
README in short
The page explains three layers: a personal wiki, a team's shared brain and agent memory. It covers installation on Mac, Windows and Linux, state capture measurements and the license.