How Conbersa Works

Conbersa has three moving parts: devices, runs, and schedules. Understand those and you can operate the whole thing, from the web app or through the MCP server.

Devices hold accounts

Every device is a physical Android phone with a stable pad_id. One platform, one account, one prompt profile that tells the agent how that account should behave — full, lurk, recovery, or recovery_exploration.

You start a run by naming the pad_ids you want. Conbersa locks those devices for the run so two jobs never touch the same account at once. If a device is already busy, the run stops with the holder's name instead of quietly double-booking it. See Device fleet & pad IDs.

Runs execute tasks

A run is one of two things:

• Agentic engagement — an agent drives a device to like, comment, follow, and search, following a saved prompt, a free-form task, or the device's own profile.
• Content upload — the fleet posts one piece of media, platform by platform. The media is pushed to the devices once; platforms run in sequence.

Every run produces a group_id and a session link: https://app.conbersa.ai/runs/{group_id}. Open it and watch each device work.

The three ways to drive an engagement run

1. prompt_id — run one saved template across every device. The most repeatable option; templates carry persona, brand, model, and step limits.
2. engagement — auto-apply each device's own prompt and persona from its profile, optionally filtered by platform. Run preview_run first to see what each device will do.
3. task — free-form natural language, with an optional persona. Good for one-offs.

max_steps

max_steps caps how many steps an agent may take. Conbersa recommends a band per engagement type:

If you leave max_steps off an engagement run, the midpoint of the recommended band is used for you.

Sessions report status

You don't need to watch a stream to know what's happening. get_session_status returns whether the group is running, completed, failed, or stopped, plus per-device statuses and failures. list_sessions shows recent runs, so an agent can find a group_id it started earlier.

Schedules repeat it

A schedule is a run plus timing: once or recurring, a timing_mode of specific (an exact time) or random (a window), a timezone, and a max_runs_per_day cap. At the fire time, Conbersa builds the same run it would have built on demand. run_schedule_now fires one immediately; stop_schedule stops the active session and re-arms or finishes the schedule. See Scheduling runs.

Content runs

Content runs post media across platforms. First you ingest the media — from a public URL or base64 bytes — which gives you a conbersa-file:// URI. Then you start a run with one group per platform, each naming its pad_ids and an optional caption. Content tools are marked experimental for now. See Posting to multiple accounts.

Next

• API keys — the credential every MCP request needs.
• MCP server overview — call all of this from an AI client.