What this is
Mangos AI can be driven from Claude Code or Codex. Instead of clicking through the app, you describe what you want and it happens. Your agent settings, your drafts, your schedule and your outreach queues are all reachable by conversation.
The point is not novelty, it is time. Distribution only compounds if you actually do it every day, and the thing most likely to stop you is context switching. If you are already in a terminal, this is a five minute detour instead of a separate app you have to remember to open.
This works through MCP (Model Context Protocol), an open standard for connecting AI tools to applications. Mangos AI runs a small MCP server inside the desktop app, bound to loopback. Your AI tool talks to that, and only that. There is no Mangos AI server in the path.
Setup is two steps and takes about a minute.
Claude Code / Codex
│
│ MCP over stdio (npx @mangos-ai/mcp)
▼
local bridge
│
│ HTTP + per-session secret
▼
Mangos AI app ── 127.0.0.1:<random port>
│
└── your agents, drafts, schedules, queues (on disk)What you need
Mangos AI, running
The connection only works while the app is open. Minimized is fine.
Claude Code or Codex
Or any other client that speaks MCP. The add command differs per tool.
Node.js 18 or newer
Used to run the bridge via npx. Nothing to install by hand.
Two steps, about a minute
-
Turn on the connection in Mangos AI
Open Mangos AI. In the left sidebar, click Connect Claude / Codex, then switch on Enable local connection.
The panel expands and the status line changes from "Off" to Listening on 127.0.0.1:<port>. That number is the port the app picked for this session. It changes every time you restart the app, which is normal and nothing you need to write down.
The connection is off until you turn it on, and stays on across restarts once you have.
-
Add Mangos to your AI tool
The same panel shows the exact command to run. Copy it with the Copy button and paste it into any terminal.
Claude Codeclaude mcp add mangos -- npx -y @mangos-ai/mcpCodexcodex mcp add mangos -- npx -y @mangos-ai/mcpRun it once. You do not need to be in any particular folder, and you do not need to run it again after app updates.
Check it worked
Open Claude Code or Codex and ask:
You should get back a list of your agents with their platform, what each one posts about, and whether its schedule is running. If you see your real agent names, you are connected.
If instead you get an error, jump to Troubleshooting.
Try it out
These are good first requests, roughly in order of how much they change.
Look around
Adjust an agent's voice
Work through your drafts
Edits you make this way count as feedback, exactly as they would in the app. If you consistently rewrite drafts to be shorter, that agent starts writing shorter drafts.
Manage the schedule
Work through your outreach queue
This is what the connection is best at. A queue of thirty DMs is tedious to edit one card at a time, and trivial to describe in a sentence.
Approving marks drafts as ready to send. It does not send them. When you are happy, either click Send approved in the app or say so explicitly:
Approving is not sending
Worth understanding once, because it is the safety model for the whole feature.
Approving marks a draft as signed off
Nothing leaves your machine. It is reversible, and it is the only thing bulk send acts
on: the app's Send approved button and the send_dm_drafts
tool both ignore anything you have not approved.
Sending is always a separate instruction
And it must name the exact drafts. There is no "send everything" tool. So "approve these" can never be mistaken for "send these", and a vague request cannot turn into a queue-wide blast.
Every send goes through the same machinery as the button in the app: your daily cap, human-like pauses between messages, and the same stop control. Nothing gets a faster or looser path just because a terminal asked for it.
What it can and cannot do
It can review, edit, approve, schedule, and send outreach
This includes agent settings, voice and topics, drafts, schedules, and your DM and email queues. Scheduled content goes back to Mangos AI for execution through its normal schedules, plan checks, and daily limits.
It cannot start an agent run or post to a social feed
There is no tool for either. It cannot make an agent go hunting for new prospects, and it cannot put a post on Twitter, LinkedIn, Threads, Reddit, Instagram or Facebook. Those stay in the app.
It cannot see your API keys
Provider keys, passwords and login tokens are stripped before anything leaves the app. If you ask Claude for your OpenAI key, it will not have it.
Tool reference
You never call these by name. You describe what you want and your AI tool picks. They are listed here so you can see exactly what the connection is able to touch.
Agents
| Tool | What it does |
|---|---|
list_agents | Every agent, with platform, persona, limits and scheduler state |
get_agent | One agent in full: config, voice profile, today's counters, recent actions, a window of its drafts |
update_agent | Change configuration, for example reply tone or a daily limit |
set_agent_voice | Replace the topics it writes about, or the keywords it searches |
set_agent_scheduler | Start or pause an agent's schedule |
Drafts awaiting review
| Tool | What it does |
|---|---|
list_pending_drafts | Posts and replies an agent has drafted, not yet scheduled |
update_pending_draft | Rewrite a draft. Counts as a learning signal |
delete_pending_draft | Discard a draft. Also a learning signal |
approve_draft | Approve it and put it on the schedule, at a time you give or the next sensible slot |
Scheduled posts
| Tool | What it does |
|---|---|
list_scheduled_posts | Everything queued to go out, with times and status |
update_scheduled_post | Rewrite the text without changing when it goes out |
reschedule_post | Move it to a different time |
retry_scheduled_post | Re-queue one that failed |
cancel_scheduled_post | Take it off the schedule |
DM outreach (Twitter and Reddit)
| Tool | What it does |
|---|---|
list_dm_drafts | The review queue: prospect, fit score, reasoning, drafted message. All DM agents at once by default |
update_dm_draft | Rewrite one DM |
bulk_update_dm_drafts | Rewrite many in one pass. This is what makes "reword all of these" fast |
approve_dm_drafts | Mark drafts ready to send. Does not send |
unapprove_dm_drafts | Put them back in review |
send_dm_drafts | Send the drafts you name. Real messages, subject to your daily cap and pacing |
skip_dm_draft | Drop a prospect from the queue |
Email outreach
| Tool | What it does |
|---|---|
list_email_drafts | The queue: recipient, subject, body, send status |
update_email_draft | Edit subject, body, and/or correct the recipient address |
bulk_update_email_drafts | Edit many in one pass |
approve_email_drafts | Mark drafts ready to send. Does not send, and does not set a time |
unapprove_email_drafts | Put them back in review |
send_email_drafts | Send the drafts you name, from your connected mailbox |
schedule_email | Commit a draft to a specific send time |
reschedule_email | Change that time |
cancel_email_schedule | Unschedule it. Stays in the queue as a draft |
skip_email_draft | Drop it from the queue |
Privacy and security
Everything stays on your machine. There is no Mangos AI server in the path and nothing about your agents is sent anywhere.
-
The connection listens on
127.0.0.1only. Other devices on your network cannot reach it, including on shared office or cafe Wi-Fi. - Every request needs a secret key that is regenerated each time the app starts.
- Turning the connection off in Settings closes it immediately.
- Quitting Mangos AI closes it too.
If you want it available only while you are actively using it, leave it off and flip it on
when you need it. The claude mcp add step does not need to be repeated.
Troubleshooting
"Could not reach the Mangos AI app"
The app is not running, or the connection is still off.
- Is Mangos AI open? Check your system tray or dock, not just the taskbar.
- Open Connect Claude / Codex in the sidebar and confirm the toggle is on and the status reads "Listening on 127.0.0.1:...".
"Mangos AI is not answering"
The app was force-quit or crashed and left stale connection details behind. Start Mangos AI again and retry. Nothing is broken and nothing is lost.
Claude says it has no Mangos tools
The claude mcp add step did not take effect.
- Run
claude mcp listand check thatmangosappears. - If it does not, run the add command again and read the output for errors.
- If it does appear but tools still are not available, restart Claude Code. MCP servers are picked up when the session starts.
"npx: command not found"
Node.js is not installed, or not on your PATH. Install it from nodejs.org (any version 18 or newer), open a new terminal, and run the add command again.
The status line says "Starting…" and stays there
Close and reopen the modal. If it still does not show a port, restart Mangos AI. This usually means another program is holding on to a resource the app needed.
Changes I make in Claude do not show in the app
They should appear within a second, on whichever screen is open. If they do not, switch to another tab and back. Your changes are saved either way; only the display is behind.
I want to disconnect
Switch off Enable local connection in the app. To remove it from Claude Code entirely:
claude mcp remove mangosFrequently asked
Do I need a paid Mangos AI plan?
No. Reading and editing works on every plan. Starting an agent's schedule and sending outreach need an active plan, the same as they do in the app. If yours has lapsed you will be told so, and everything else keeps working.
Can Claude send messages without asking me?
It has to name the exact drafts, and it is told to confirm with you first and never to treat "approve these" as permission to send. Your daily cap and pacing apply either way, so the worst case is bounded by the same limits as the app's own Send button. If you would rather keep sending entirely in your own hands, approve from the terminal and click Send approved in the app.
What happens if a send fails?
The draft stays in your queue with a "Send failed" chip and the reason, exactly as it would in the app. Nothing is silently dropped.
Does this use my AI credits?
Not Mangos AI's. Your Claude Code or Codex usage is billed by whoever provides that tool. Mangos AI agents keep using whatever model you configured for them.
Can I use this with other AI tools?
Any tool that speaks MCP can connect the same way. The exact add command differs per
tool. Point it at npx -y @mangos-ai/mcp.
Can two tools connect at once?
Yes. Claude Code and Codex can both be connected, and both will see the same live data.
Does it work if Mangos AI is minimized?
Yes. It only needs the app to be running, not visible.
Can someone else on my network connect to it?
No. The connection is bound to your machine's loopback interface and is not reachable from anywhere else.