log

log runs your social accounts from the command line, or from your AI agent. It reads and acts on X, Reddit, Facebook and Discord (conversations, replies, DMs, Marketplace listings and groups), and signs in to LinkedIn. It's one command, log, and one MCP server, logMcp. It runs on your own machine with your own accounts, and it's free and open source.

Why use it

• Find the people already asking for what you do. Read your DMs, the replies under your posts, groups and Marketplace, and see who's waiting on you and what they want.
• Answer them in your own voice. Drafts are written from how you already talk to each person, and nothing is sent until you say so.
• Run many accounts without paying for a SaaS. Your accounts and sessions stay in ~/.logCli on your machine.
• Let your agent do the legwork. Claude Code, Codex or Cursor runs the same commands through the MCP server.

Install

macOS and Linux:

curl -fsSL https://log.listeningkit.com/install.sh | sh

Windows (PowerShell):

irm https://log.listeningkit.com/install.ps1 | iex

This installs the log and logMcp commands, and Python with everything the platform engines need. It writes your settings to ~/.logCli/.env. Already have Bun? Then bun add -g @listeningkit/log installs the commands on their own. More: docs/install.md.

Get started

1. Add an account and sign it in. Export the account's cookies from your browser with a cookie extension, then paste them when log asks. They're typed into a masked prompt and never passed as flags.

log localAccount add --platform twitter --account <name>
log twitter cookies save --account <name>
log twitter session check --account <name>

The last line asks the platform itself whether the session is signed in.

2. Use it. For example:

log twitter dms --account <name>           # who's waiting on you, and what they asked
log twitter session check --account <name>
log --help                                 # every command

3. Hand it to your agent.

claude mcp add log -- logMcp

Then ask in plain words: "read my X DMs and tell me who's waiting on me". Codex, Cursor and other clients: docs/mcp.md.

How we use it

We run log every day from Claude Code. The agent loads log's guide, then works through the MCP tools:

• reading the X inbox and following up with everyone who's waiting
• answering the people who engaged with a post
• DMing the people who liked one
• keeping Marketplace listings live across local accounts
• listening to local Facebook groups for people asking for a service

Each of those is a use case, written step by step: what to run, and what tells you each step worked. So you can run the same play.

Getting the most out of it

• Start from a use case and follow it, rather than calling commands one by one.
• Nothing acts until you say so. Every command that changes something on a platform is a dry run unless you add --live. Over MCP it's a separate …Live tool that needs your confirmation.
• Keep it human. Each account has its own daily limits, pacing and local waking hours, and log sticks to them.
• Many accounts? Keep them in Twenty, an open-source CRM. Every machine, agent and teammate then sees the same accounts, and you can edit them by hand.

Documentation

The full docs are at log.listeningkit.com/docs, built from docs/:

Where each page sits is set in docs/site.json, and a commit check refuses a page that isn't placed.

Contributing

git clone https://github.com/matthewdonsemail-lab/log
cd log
bun install    # also installs the git hooks
bun link       # puts this checkout's log and logMcp on your PATH

Every commit and push runs the same checks we do (lefthook). They check that:

• the docs and the reference match the code
• the naming and layout are right
• there are no comments in code
• the tests pass

The rules agents work by are in AGENTS.md.

License

MIT