magpie-local

A local memory store for magpie. Your captures land in a SQLite file on your own machine, you can search them, and Claude can read them over MCP.

No account, no key to obtain, no network. The pages you save never leave the computer they were saved on.

Documentation: magpie.anonalabs.com/local

  magpie extension ──POST──▶  127.0.0.1  ──▶  ~/.magpie/memories.db
                                   ▲
  Claude Code / Desktop ──MCP──────┤   search · get_page · recent
                                   │
  your browser ────────────────────┘   magpie-local ui

Install

Node 22.5 or newer, which is the only requirement: the database is Node's own node:sqlite, so there is nothing to compile and no platform-specific binary.

npm install -g @anona-labs/magpie-local
magpie-local install        # runs it now, and from every login onwards

The package is scoped; the command it installs is not. Or without installing anything permanent: npx @anona-labs/magpie-local start runs it in the background until you reboot, and ... serve runs it in the terminal where you can watch it.

It prints a token the first time. Paste that into magpie: Settings → This machine → Token. It is in ~/.magpie/token if you need it again, or run magpie-local token.

Then let Claude read it:

claude mcp add --scope user magpie -- magpie-local mcp

--scope user matters: the default is local, which registers the server for the directory you happened to be in and nowhere else.

The commands

| | | |---|---| | magpie-local | what is in the store, and whether it is running | | magpie-local install | start it now, and with the computer from now on | | magpie-local uninstall | stop doing that | | magpie-local start | run it in the background until the next reboot | | magpie-local stop / restart | | | magpie-local serve | run it here, in the foreground | | magpie-local ui | open it in a browser: search, read, forget | | magpie-local search <query> | search the store from the terminal | | magpie-local token | print the token, for pasting into magpie | | magpie-local mcp | speak MCP on stdio; Claude starts this itself |

Published as @anona-labs/magpie-local. anona on npm is somebody else's package; Anona Labs' own packages are the @anona-labs scope on npm and anona on PyPI.

magpie-local on its own is the one that answers everything: whether the service is up and for how long, how much is stored, how much of it is embedded, which spaces and what kinds of thing they came from, a fortnight of capture as a sparkline, and the newest thing you saved.

It is read from your own file and printed on your own terminal. magpie has no telemetry: none of it is sent anywhere, by that command or by anything else.

magpie-local ui opens http://127.0.0.1:7777, starting the store first if it is not already up. The page shows what is stored, searches it, reads a whole page back, and forgets one you no longer want.

The token reaches that page in the # of the address, which a browser never sends to a server and never writes to a log; the page keeps it, so every visit after the first is just the address. Everything the page needs is in the document itself, so no font, script or stylesheet is fetched from anywhere and it works with the network down. The page is served without a token because it holds no data: every number on it comes from a call the token guards.

start writes a pid file and a log in ~/.magpie/, and detaches, so closing the terminal does not take the store with it. It does not survive a reboot.

install is the one that does. It writes your platform's own service file and enables it, so the store is running whenever you are logged in:

| | | |---|---| | Linux | ~/.config/systemd/user/magpie-local.service, enabled with systemctl --user | | macOS | ~/Library/LaunchAgents/com.anonalabs.magpie-local.plist, loaded with launchctl | | Windows | a .cmd in the Startup folder |

It prints the file it wrote and the commands it ran, so nothing happens to your machine that you cannot read first. magpie-local uninstall removes it. On Linux it starts at login and stops at logout; loginctl enable-linger $USER makes it run without you logged in at all.

serve and mcp are separate processes on purpose. Claude starts the MCP one itself and owns its lifetime; serve runs while your browser does. They share the file, which is what SQLite is for.

What Claude gets

| Tool | For | |---|---| | search_memories | keyword and semantic search across everything you saved | | get_page | the whole text of one page, by id or URL | | recent_memories | what you saved lately |

All three are read-only. A tool that can write is a tool a prompt injection in a web page can aim at your notes, and the writing surface already exists in the extension behind a key you press.

There is deliberately no ask tool and no model inside the server. In an MCP setup the generation half of retrieval-augmented generation is the client: Claude is doing the reasoning, so this server's whole job is to find the right text and hand it over.

How search works

Two retrievers over the same file, fused with Reciprocal Rank Fusion:

• FTS5 with the Porter stemmer, so "compaction" finds a page that says "compact", and each word you type is matched as a word rather than as query syntax;
• vectors from bge-small-en-v1.5, 384 dimensions, embedded on the CPU in this process.

The model is about 130MB and downloads once, in the background. Nothing waits for it: captures are accepted and keyword search answers from the first second, and semantic results start appearing on their own when the download finishes. Every result says which retriever found it.

Vectors are scanned exactly rather than through an approximate index. At the scale one person reaches, roughly 70k chunks a year, that is ~107MB and tens of milliseconds, and exact means no index to rebuild and no recall cliff.

What it stores

One row per page: the URL, the title, when you saved it, the summary magpie wrote, and the full source text. The extension sends a summary to cloud providers because they bill for what they extract; on your own disk there is no bill, so the source is kept and indexed while the summary stays as the headline.

Re-saving a page replaces it rather than storing it twice.

~/.magpie/
  memories.db     pages, chunks, the text index and the vectors, one file
  token           the credential, mode 600

magpie-local says what is in it. sqlite3 ~/.magpie/memories.db opens it, because it is an ordinary SQLite file and it is yours.

Security

Any page you visit can make a request to 127.0.0.1. A store holding the full text of everything you have read is a bigger prize than one holding summaries, so three things guard it:

• a token on every request except /health;
• CORS headers only for extension origins, so a web page that does get a request through still cannot read the answer;
• a Host check, because DNS rebinding turns a hostile name into 127.0.0.1 and the browser sends the request anyway.

The database is not encrypted. Its directory is mode 700 and the token file 600, which is the same protection your browser profile has beside it. If that is not enough for what you read, it should not be captured.

Limits, deliberately

This is one machine. No sync, no sharing, no phone. It searches what you saved; it does not consolidate facts, track what supersedes what, or reason across your memories. That is Anona Memory, and it is a different product that happens to also store text.

Configuration

| Variable | Default | |---|---| | MAGPIE_HOME | ~/.magpie | | MAGPIE_PORT | 7777 | | MAGPIE_EMBED_MODEL | Xenova/bge-small-en-v1.5 |

Changing the model invalidates every vector. Nothing breaks and nothing has to be run: each vector records the model that wrote it, search ignores the ones that do not match, and they are re-embedded in the background.

Tests

npm test

The store, the search, the fusion and every guard above. They use node:sqlite and the package's own source only, so they run without the embedding model being present.

MIT.