For people who run many X accounts
Auto reply for X (Twitter) from many accounts, paced
Flottra posts replies for many X (Twitter) accounts with text you or an AI agent supply. Pick the post, give the text, and the desktop app replies from the account’s own browser on your computer, through its proxy, marked done only when X confirms. Pods stagger 3 to 6 accounts on one post; agents schedule many through the MCP server.
Updated
What it does
A reply engine for many accounts, not a reply bot for one
You choose the post and the words. The desktop app posts the reply from the account you name, in that account's own browser window on your own computer, through that account's own proxy, and marks it done only when X confirms it. Across a pod, or from an AI agent, that is auto reply at fleet scale with the pacing built in.
Search for twitter auto reply and page one is two kinds of tool: a workflow you build yourself on n8n, Make, IFTTT or a GitHub script, answering from one account through the official API, or a direct message autoresponder like the quick replies X offers business accounts. Neither is built for someone with fifteen accounts who wants one post answered from six of them.
Flottra is the third kind. Give it a post on X and the reply text; the desktop app posts it from that account's own browser profile on your computer, through its own proxy. A pod does it for 3 to 6 of your accounts on one post, staggered. An AI agent through the MCP server can write a different reply per account and schedule each as its own task, gate and caps applied.
What it is not: it does not read your mentions, and in the panel it does not pick the post or write the reply. That part is the engagement autopilot, set up through the MCP server: it searches topics you list and replies in the account's persona voice at a small daily budget. It sits beside auto follow and the rest of the feature set.
- 10:52@jpegvaultfeed_reply to @solrpc_dev: 'floor is down and my bid is still up there'queued
- 10:31@wifhat_maxifeed_reply to @solrpc_dev: 'heart monitor chart, somehow bullish'running
- 10:14@degenrouterfeed_reply to @solrpc_dev: 'that nonce bug again'done
- 09:47@nodefarmerfeed_reply to @solrpc_dev: CreateTweet not confirmedfailed
Four feed_reply tasks an agent scheduled through the MCP server, as the task log shows them. Hand set.
How it works
Pick the post, supply the text, the app does the rest
Three steps, and the third one is the app's. The reply goes out the way X's own front end sends one, and it only counts when X says it landed.
- 01
Pick the post
Paste the URL of the post on X you want answered. In a pod it is one post for the whole group. Through the MCP server an agent passes a post with each task, so every account can answer a different one. - 02
Supply the text
Up to 280 characters, posted as written. You type it in the pod form, or an agent drafts it for each account from that account’s persona and passes it with the task. The app never calls a language model. - 03
The app replies from the account’s own browser
The desktop app opens the post in that account’s browser profile through its proxy, pauses the way a reader would, posts the reply, and marks the task done only when X itself confirms it. The reply’s URL is kept on the task.
Two paths, in a fixed order: X's own CreateTweet call, fired from inside the page the way X's front end fires it, and, only if that is unavailable, the reply composer, typed at a human rhythm. Done means X answered with the id of the new reply, which the log links to; a reply typed but never confirmed is marked failed, with the reason on the task. The app runs one task at a time, one browser session after another; if it is closed or the laptop sleeps, the reply waits as pending and runs when it is back. The app is Windows today.
- 280
- characters per reply, posted as written
- 5
- replies per account per hour when an agent schedules through the MCP server
- 3 to 6
- accounts in a pod, one post, one text, staggered
- 1
- proxy per account; without one the app refuses to open a browser
Voice
Ten accounts, ten different replies
The text is the one part Flottra leaves to you or to an agent, on purpose.
The quickest way to tie ten accounts together is one sentence posted from all of them under one post. X's authenticity policy calls it content spam, and a synonym swap still reads as one hand. The reply bots guide covers what gets caught. A reply survives when it is specific to the post, short, without a link, and different for every account.
Each account can carry a persona: a bio, a voice, what it talks about and avoids, how long it writes, example posts. Write one by hand, or have the panel draft it with your own Anthropic key, then edit it. For replies a persona is read in two places: by an AI agent through the MCP server before it drafts a reply for that account, and by the engagement autopilot when it writes one itself. The AI personas page has the fields.
A pod is the opposite tool: one run posts one text from every member, fine for a two word answer everyone would give anyway, wrong when the replies should read as different people. For different words per account, schedule separate tasks or let an agent write each one.
Pacing
Gated by account age, staggered inside a pod, capped per hour
Every check happens when the task is created. That is also the honest place to say what is checked and what is not.
A new account cannot reply on day one. The warming gate runs on account age: days 1 to 3, likes and profile views only, 3 a day; days 4 to 7 add follows, 6 a day; days 8 to 14 add replies, 12 a day; from day 15 the account is open. It applies to everything scheduled through pods, identity and the MCP server, and refuses a reply the phase does not allow or that would pass the day's cap. The account warming page has the ramp in full.
A pod never fires its members together: the order is shuffled, the first member waits 20 seconds to 2 minutes, each one after that a further 1 to 4 minutes at random, so six accounts on one post are spread across minutes in an order that carries no signal. Suspended, locked, retired or gate refused members are skipped.
Through the MCP server there is also a per account hourly cap: 5 replies an hour by default, editable in the workspace settings. A bulk call stops at 50 accounts, 5 tasks per account and 100 tasks, spread over a random window, and previews until the agent confirms.
Two things pacing does not do. A reply you schedule has no quiet hours and no per account clock: a reply scheduled for three in the morning on the account's own time goes out then, so schedule for its daytime, or have the agent do it. And the checks run when a task is created, not when it runs: pods apply the warming gate, the MCP server the gate and the hourly caps; the app runs what is queued, in order.
Beta status
What runs today, and what does not yet
Kept apart on purpose, so a tester knows exactly what to expect before the first task.
Today, on the hosted beta
- Replies with text you supply, from a pod: 3 to 6 of your accounts, one post, one text, staggered. Pro and Team plans.
- Replies with text an agent supplies, through the MCP server.
reply,smart_replyandfeed_replyall run on the desktop app, each with its own post and text, gate and caps applied. - The engagement autopilot, set up through the MCP server. The account searches the topics you list, and the server queues likes, follows and replies written in the persona's voice: 4 replies a day by default and at most 10, each at most 140 characters, with no link, @handle, hashtag or question. It needs desktop app 0.1.48 or newer, a persona and your own AI key.
- The reply itself. Posted from the account's own browser on your computer through its proxy, confirmed by X's own response, the reply URL kept on the task.
- Personas per account. Written by hand or drafted in the panel with your own Anthropic key, as the brief every reply is written from.
Not on the hosted beta yet
- Replying to mentions on its own: mentions and replies on the account's own posts, the opted in kind in X's rules, with nobody passing each one in.
- A panel page for the autopilot, and for a single typed reply. Outside a pod, replies are scheduled through the MCP server.
- Photos on replies. A reply is text.
When one of those moves to the desktop app, this page will say so.
The rules
What X says about automated replies, quoted
X's automation rules say more about replies than about any other action, and the tools on page one do not quote them. From the page marked updated April 2026:
You "may not follow or unfollow X accounts in a bulk, aggressive, or indiscriminate manner". You "may not like posts or hide replies in an automated manner". Unsolicited automated direct messages are not permitted. Scripting the X website "may result in the permanent suspension of your account".
The same page, quoted at length in the reply bots guide, adds two lines on replies: automated replies to posts "based on keyword searches alone is not permitted", and "the deployment or operation of any AI reply bot requires prior written and explicit approval from X". Neither a language model nor an agent connected to this one comes with that approval. The engagement autopilot finds posts by topic search and writes the reply with a language model, which is the pattern those two lines name. Nothing in Flottra changes that, so decide with the quote in front of you.
Flottra is browser automation and does not claim those rules do not apply to it. What it does is refuse the shape the page names: one reply per task per account, a warming gate for young accounts, a stagger inside a pod, an hourly cap through the MCP server, one proxy per account. That reduces the rate at which accounts are locked or suspended. It does not make it zero, and nobody can.
The long versions: reply bots for what gets caught, rate limits for the one published daily figure and what a cautious operator runs, and shadowban for how to see, from a logged out browser, whether your replies sit behind show more replies.
Compared
A workflow, an autoresponder, or a reply engine
The three kinds of tool on page one for this search solve three different problems. Here they are on the points that matter once you run more than one account.
| A workflow you build (n8n, Make, IFTTT) | A DM autoresponder | Flottra | |
|---|---|---|---|
| Accounts | One per workflow, with its own API key | One business account | Many in one workspace: 2 on the free plan, 10, 40 or 150 on the paid ones |
| Where it runs | The workflow vendor’s cloud, or a server you host | Inside X, in direct messages | The desktop app on your own computer, one browser profile per account |
| Whose IP | The vendor’s or your server’s, for every account you connect | Not applicable | Your own proxy, one per account, residential, mobile or ISP |
| Who writes the text | A template in the workflow, or a model you wire in | You, as saved replies | You, the pod form, an AI agent through the MCP server, or the engagement autopilot on the server; never the app itself |
| Pacing | Whatever the trigger fires, whenever it fires | Instant, which is the point of it | A warming gate by account age, a shuffled stagger in a pod, 5 replies an hour per account through the MCP server |
| When X changes its design | Unaffected by the website; exposed to API tier and price changes | Unaffected | A fix ships as an app update; the app checks for updates while it runs and installs them when idle |
Each of the first two is the right tool for its own job: one account thanking the people who mention it, a shop answering a direct message at night. Flottra is for the third job: many accounts answering chosen posts, each from its own browser and address, at a pace a person could have kept. It sits next to auto follow and account warming, and starts at two accounts free, no card.
Questions
Questions people ask about this
- How to set an auto reply message?
- In Flottra you set the message per reply, not per account. In a pod you paste the post you want answered and type one message of up to 280 characters; the desktop app posts it from 3 to 6 of your accounts, each from its own browser on your computer, staggered. An AI agent connected through the MCP server can write a different message for each account and schedule each one as its own task. A stored message that answers mentions on its own does not run on the hosted beta.
- Is there a Twitter bot that can automatically reply to tweets?
- Yes, in three shapes. Workflows built on n8n, Make or IFTTT reply from one account through the official API when a trigger fires. Business accounts get X’s own quick replies inside direct messages. Flottra replies from many accounts to a post you choose, with text you or an AI agent supply, from a real browser on your own computer through each account’s own proxy. Set up through the MCP server, its engagement autopilot can also search topics you list and reply in the account’s persona voice at a small daily budget. It does not answer mentions.
- Does Tweetbot still work?
- No, as far as we know. Tweetbot was a third-party Twitter client made by Tapbots, an app for using your own account rather than a reply bot, and it shut down in January 2023 after X cut third-party apps off from its API. Flottra is not a client and does not use the API. The desktop app drives a real browser window per account on your own computer, so it does not depend on the API access that third-party clients lost.
- How to set up auto tweets?
- Scheduled posting is a separate feature from replies, and it runs today. In Flottra you write a post in the panel, or let the account’s persona write it, choose the accounts and a time, and the desktop app publishes it from each account’s own browser. A posting routine can also write and publish through the day. The scheduling page has the form, the routine and the limits. Replies work with text you supply or, through the MCP server, with text the engagement autopilot writes.
- Can Flottra reply to my mentions automatically?
- Not on the hosted beta. Replying to mentions and to replies on an account’s own posts exists in the product as the engage_notifications task, but it runs on a worker lane that is not deployed for beta workspaces, so nothing reads your notifications today. What runs is a reply to a post you choose, with text you, a pod or an AI agent supply, and the engagement autopilot’s replies to posts it finds by topic search. When mention replies move to the desktop app, this page will say so.
- Will X shadowban an account that replies a lot?
- It can limit what your replies reach. X’s enforcement page lists downranking a post in replies among its options, which is what people mean by a reply shadowban: the reply is still there, behind "show more replies". The patterns that draw it are the same text from several accounts, links in replies, and a high volume from a young account. Flottra paces replies and gates young accounts, which lowers the rate. It does not make it zero, and nobody can.
- Do I need a proxy for every account?
- Yes. The desktop app refuses to open a browser for an account that has no proxy, so a reply from an account without one never runs. One proxy per account is the rule the product recommends and the MCP prompts enforce, and it should be residential, mobile or ISP, the only types the panel accepts. Flottra does not sell proxies. You add your own in the panel, one at a time or in bulk, and assign one to each account.