CLI and MCP
Work the queue from a terminal or an agent session.
EarShots ships a command line tool and an MCP server that read the same records the app shows. List mentions, open one post with its reasoning and citations, and store a reply draft without leaving the shell, or hand the same tools to an agent already doing the work.
$ earshots insights
questions 12, friction 5, buying signals 3, praise 9, muted 41
$ earshots mentions --track answer --limit 10
10 posts, newest first, each with an id, channel, and status
$ earshots draft m_01h9 --body
draft saved for review. nothing was posted
Tools an agent can call over MCP
- list_mentions
- get_mention
- get_insights
- get_sentiment
- get_launch_reactions
- save_reply_draft
save_reply_draft stores a draft against a mention for a person to review. No tool sends a reply, posts a comment, or messages an author.
What you see
The CLI
Commands across the queue: earshots insights, mentions, show, sources, buzz, stats, launch, and draft. Output is plain text by default and structured with --json.
The MCP server
Tools that read mentions, insights, sentiment, Context Buzz, launch reactions, and competitor signal, plus save_reply_draft for storing a draft against a post.
The guardrail
No command and no tool posts a reply, opens a thread, or messages an author. Writing to a network is not part of the surface.
How it works
Sign in with earshots login: the browser opens, you approve once, and the token is stored locally and refreshed on its own. Commands then run against the active workspace, the same one the app has open, and you switch with earshots workspaces.
FAQ
Common questions
How do I install the CLI?
npm install -g @earshotsdev/cli, then run earshots login. It needs Node 20 or newer.
Does an agent need its own key?
An agent session works with the same workspace access as the signed-in user, so plan limits and review rules apply to it the same way.
Can an agent send the reply?
No. The tools read data and store drafts. Sending stays with a person in the app.
Are any tools limited by plan?
get_launch_reactions is on Growth and Enterprise. get_competitor_signal follows the rival count on the workspace plan. The rest follow the workspace plan the same way the app does.
Point a terminal or an agent at the queue
Install the CLI, run earshots login, and connect the MCP server to the workspace you work in.