What is a social listening agent?

Software that reads what people post about you and acts on it without being asked each time: deciding what is relevant, what the author wants, who should see it and whether to reply. In FishOut that is two things — its own agent rules, and any AI agent you connect over MCP.

Do FishOut’s agents post replies?

No. They label, assign, set status, notify and draft. Every reply is posted by a person, from their own account.

Can I use my own AI agent instead?

Yes. Connect Claude, ChatGPT, Codex, OpenClaw, Hermes Agent or another MCP client to FishOut’s server. It can read and, if you grant it, act on your mentions. That works on every plan, including Free.

Does FishOut track what ChatGPT says about my brand?

No. FishOut reads the conversations people have on social platforms and in the press — the input AI assistants often cite, not their output. Measuring what an AI answers about you is a different kind of tool.

Which should I start with?

Start with FishOut’s built-in views and an alert. Add an agent rule when you find yourself doing the same thing to every lead. Connect your own assistant when you want to ask questions the views do not answer.

Social listening agents

Two ways to put an AI agent on your mentions

FishOut is a social listening agent in its own right: it reads every post for relevance and intent, decides whether it is worth a reply, and runs your rules on the result. And it is an MCP server, so the agent you already use — Claude, ChatGPT, Codex, OpenClaw — can do the asking.

Start free Connect your own agent

[01] FISHOUT’S OWN AGENTS/ BUILT IN

Every post is read before you see it

Each mention gets a relevance from 0 to 100 with the reasons, labels for what the author wants — Asking for recommendations, Comparison, Has a problem — and a Reply call: Reply now, Reply carefully, Watch or Skip. Noise goes to Filtered with a reason and never counts toward your plan.

Reply callReddit · Hacker News · YouTube

Reply nowFresh, asking for help, nobody has answered

Reply, carefullyWorth it, if you lead with help

WatchAlready answered, or written by a vendor

SkipPromotion, bots, or a link would read as spam

Hot leads9 runs / 24h

When

Every triaged mention

Reply call is Reply now

Subject is A topic

Then

Draft a reply

Assign the mention to Mark

Post to Slack #inbound

Then your rules act on it

An agent rule reads as a sentence: when a mention is triaged, and only if these conditions hold, then label it, assign it, set its status, draft a reply, or send it to Slack, email or a webhook. Every run is logged. Agents are on Plus and up; see the docs.

[02] YOUR AGENT, OVER MCP/ ANY CLIENT

Or ask the agent you already use

FishOut’s MCP server gives an assistant the same tools as the API, within the projects and permissions you approve. It can cross-check a mention against your code, file an issue with its own GitHub access, or write the weekly note — things a rule cannot. All fifteen client guides.

OpenClaw Hermes Agent

Grok

Codex · terminalfishout · mcp

Which of today’s Reply now posts are about the API, and is any of it a bug?

list_mentionsread-onlyarguments

{
  "filter": {
    "replyCalls": [
      "reply_now"
    ],
    "query": "API",
    "from": "2026-10-20T00:00:00Z"
  },
  "limit": 10
}

get_mentionread-onlyarguments

{
  "mentionId": "mnt_…"
}

emberline/workerissue #183h

Webhook events arrive out of order after a retry

Reply nowHas a problem

Hacker Newsoakhollow5h

Their API docs are the best part of the product

Reply nowPraise

Two. The GitHub issue is a real bug: a retried webhook is sent with its original timestamp, so it lands before newer events. The HN comment is praise; a thank-you is enough. Want me to open a fix branch?

[03] WHICH TO USE WHEN/ SIDE BY SIDE

Rules for routine, your agent for judgement

Most teams end up with both: rules so nothing is missed, an assistant for the questions.

FishOut agent rules compared with your own agent over MCP
	FishOut agent rules	Your agent over MCP
When it runs	On every mention, as it is triaged, or on a clock	When you ask, or when your agent’s scheduler runs it
What it decides	Fixed rules: conditions on relevance, label, Reply call, subject…	Anything you can phrase: group, compare, explain, cross-check your code
What it can do	Label, assign, set status, draft, Slack, email, webhook	Any FishOut tool you granted, plus its other connections (GitHub, Linear…)
Who pays for the model	Included in your FishOut plan	You, through your assistant’s own plan
Plan	Plus and up	Every plan, including Free
Best for	Routine: never miss a lead, route every bug	Judgement: today’s five, this week’s themes, the launch thread

[04] PLAYBOOKS/ 8 PROMPTS

Jobs for your agent

Each is a prompt to paste, with the tools it calls and what it will not do. All playbooks.

Daily buyer-intent digest Top 5 posts to reply to today Competitor complaints summary Hacker News launch watch Public bug reports to issues Weekly share-of-voice report Keyword noise audit Reply on Reddit without getting banned

[05] GUARDRAILS/ BOTH KINDS

Post for you

Neither kind of agent can. FishOut has no posting tool and its rules stop at a draft. Automated posting is how accounts get banned on Reddit and Hacker News.

Have more access than it needs

Approve a read-only grant for an assistant that only answers questions. The server checks every call against that grant and your current membership.

Change things without asking

Every MCP tool carries hints. Archiving, editing a keyword or teaching a correction is marked destructive, so a well-behaved client asks first.

Outlive the person who connected it

Remove someone from a project and their assistant loses it too. Disconnect any assistant under Settings → API & MCP.

[06] WHERE YOU CONTROL IT/ SETTINGS

Every connected agent, in one list

Settings → API & MCP shows which assistant each person connected, the grant it holds and how much it is used. Disconnect one and it stops at the next call.

API & MCP · connected clients3 connected

Mark — Claude Code

readseen 4m ago

214

calls / 7d

Disconnect

Devon — Cursor

read · writeseen 15h ago

calls / 7d

Disconnect

Priya — Claude

readseen 2d ago

calls / 7d

Disconnect

Two ways to put an AI agent on your mentions

Every post is read before you see it

Then your rules act on it

Or ask the agent you already use

Rules for routine, your agent for judgement

Jobs for your agent

Post for you

Have more access than it needs

Change things without asking

Outlive the person who connected it

Every connected agent, in one list

Questions

What is a social listening agent?

Do FishOut’s agents post replies?

Can I use my own AI agent instead?

Does FishOut track what ChatGPT says about my brand?

Which should I start with?