What the agent can do
What an AI agent can do in Multogram over MCP: adding and checking accounts, proxies, templates, mode settings, tasks and modules, logs, the Modules Constructor. What the Free, Pro and Max plans unlock, action confirmations and limits.
The agent works through tools — separate actions that Multogram performs. The tools are grouped, and the agent gets only those your plan unlocks. The list follows the application by itself: signing in, signing out and changing the plan change the set of tools right away, with no need to reconnect the client.
Capabilities by plan
| What the agent can do | Free | Pro | Max |
|---|---|---|---|
| Knowledge — search and read this documentation, describe the application and its modes, and what is available | ✓ | ✓ | ✓ |
| Viewing — tasks, their progress and logs; accounts, proxies, templates; mode settings; what changed | ✓ | ✓ | ✓ |
| Assistance — open a section, show a notice, open the Add Account window, prepare a support request | ✓ | ✓ | ✓ |
| Accounts and proxies — add accounts (one by one or a whole folder), check them, assign proxies; add, check and remove proxies; create a template | ✓ | ✓ | ✓ |
| Mode settings — fill in the fields of a mode's section, the message text included | — | ✓ | ✓ |
| Tasks and modules — create, start, pause, stop and remove a task; run a module | — | ✓ | ✓ |
| Modules Constructor — build and edit modules: scenario, functions, form fields, test runs | — | ✓ | ✓ |
| Mass actions — check accounts, ask @SpamBot about limits, take proxies away for many accounts at once | — | — | ✓ |
On Free the agent does what you do by hand in the Accounts and Proxies sections: for example, "add the accounts from the folder D:\Sessions and check them" or "add these 50 proxies and give each account its own".
The plan unlocks tools for the agent, not modes. The agent sees in advance which modes your account can use, and for a locked one it names the plan or the purchase that unlocks it.
Tools
| Group | Tool | What it does |
|---|---|---|
| Knowledge | docs_search, docs_read | Search the documentation and read a page or one of its sections |
| Knowledge | app_overview | An overview in one call: the plan and what it opens for the agent, modes, accounts and proxies by status, tasks, the connection to the server, hints for a new installation and known problems |
| Knowledge | tasks_modes | Modes: what each one does, whether it is available, how to unlock it and whether its settings are filled in |
| Viewing | tasks_list, tasks_get | The task list and one task: status, schedule, threads, the result of its latest run and why it did not start |
| Viewing | tasks_logs, tasks_wait | A task's log (including runs from before Multogram restarted) and waiting for it to finish |
| Viewing | accounts_list, proxies_list | Accounts and proxies: statuses as the tables name them, filters, a summary. No session files or passwords |
| Viewing | templates_list, templates_get | Message templates and the text of one template |
| Viewing | settings_get | A mode's settings: fields, values, required and empty ones |
| Viewing | interface_get_state, events_since | What is open in the window; what changed — task states, account statuses, proxy failures |
| Assistance | interface_switch_tab | Open a section of the application |
| Assistance | interface_notify | Show a short notice in the window, marked as coming from the agent |
| Assistance | accounts_authorize, dialog_wait | Open the Add Account window (you type the SMS code) and wait for how it ended |
| Assistance | support_ticket | Open the support form with a prepared text. You send it |
| Accounts | accounts_add | Add accounts: a folder of .session + .json pairs, separate files or a list. The phone comes from the file name; phones already added and tdata are skipped with the reason |
| Accounts | accounts_check | Check accounts — sign in to each and refresh its row, like the table's Check accounts action |
| Accounts | accounts_assign_proxies | Give accounts proxies: one per account (or in turn), working ones only, without replacing those already assigned |
| Accounts | accounts_update, accounts_remove | Change an account row's name or proxy; remove accounts (session files stay on disk) |
| Proxies | proxies_add, proxies_check, proxies_remove | Add proxies (as text in any common form), check them, remove them |
| Templates | templates_create | Create a template: formatted text and attachments |
| Settings | settings_update | Fill in a mode's settings, the message included: text, formatting, attachments or a template |
| Tasks | tasks_create | Create and start a task. The agent can first check the launch (dry_run): what will run, from which accounts and with which settings — creating nothing |
| Tasks | tasks_start, tasks_stop, tasks_remove | Start or resume, pause or stop, remove a task |
| Modules | modules_list, modules_run | Installed modules and their forms; running a module as a task |
| Mass | accounts_bulk | A mass action on accounts: check, limits, taking proxies away |
| Constructor | constructor_status, constructor_open, constructor_tools | Whether the constructor is open and on which module, opening it, finding the constructor's tools |
| Constructor | constructor tools | While the constructor is open: scenario, functions, actions, validation, test runs, saving the module; the others (variables, form fields, tabs, logs) on request. Close the window and they are gone |
Two more tools are always there, even when the application is not running: get_access_status — what is
available and what to do when there is no access — and get_guide, the built-in guide for the agent.
What the agent cannot reach
- Secrets. The agent neither sees nor fills API keys, passwords or other secret fields — you enter them.
- Session files and passwords. The agent says where to add accounts from, but receives neither the files' contents, their paths nor proxy passwords.
- The SMS code and signing in to Telegram. The agent can open the Add Account window, but you type the code.
- Sending requests. The agent prepares the text; you send the request.
- Chains, files and databases in a mode's settings, picking folders — you fill those in the section itself.
- The protected module field. The agent does not change or remove the accounts field of a module's form — only its value.
Confirmations
Changes go through you. Multogram shows the Approve an AI agent's action window: who asks and what exactly will happen — the mode, the accounts, the settings and the message text, the proxy addresses.
| What the agent does | When you are asked |
|---|---|
| Adds and checks accounts and proxies, assigns proxies, changes settings | The first time in the agent's session |
| Removes accounts, proxies or a task | Every time |
| Creates and starts a task, runs a module, starts a mass action | Every time |
Checks a launch (dry_run), pauses or stops a task | Never |
| A module test in the constructor, removing in a module | The first time in the session; probes that only read are not asked about |
A module test runs for real — messages are sent from the accounts you named.
| Button | What it means |
|---|---|
| Allow | Perform this action once |
| Allow for this session | The agent may repeat this action without asking until it disconnects |
| Decline | Do not perform it. The agent is refused and asks what you want instead |
If you do not answer for about two minutes, the window closes and the action is not performed. Your AI client's own confirmations are an extra layer: the Multogram window is shown either way.
Limits
| Limit | Free | Pro | Max |
|---|---|---|---|
| Agent sessions working at once | 1 | 2 | 4 |
| Calls per minute per session | 20 | 60 | 120 |
| Accounts or proxies in one call | 100 | 250 | 500 |
The agent adds a large folder in several calls. Only a working session takes a place in the session limit: the agent is performing an action or performed one within the last five minutes. An open but idle client holds no place.
When there is no free place, Multogram shows the AI session limit window: which agent is working now and which client started it. Disconnect stops its current action and frees the place — the agent can connect again later. Keep it — the waiting agent starts once the current one has been idle for five minutes.
Frequently asked questions
Can the agent start a task?
Yes, on Pro and Max: the agent fills in the mode's settings, checks the launch and shows you a summary, and you approve the launch in the Multogram window. Every launch, like yours, is checked by the Multogram server.
What can the agent do on Free?
Everything you do by hand in the Accounts and Proxies sections: add accounts from a folder, check them, add proxies and hand them out to accounts, create a template. Plus answering from the documentation, showing tasks and logs and helping you find your way around.
Does the agent see my accounts and sessions?
The agent sees the accounts database as a table: phone, name, status, country, proxy address and whether the account has a session. It never receives session files, their paths, passwords (proxy passwords included) or codes.
Updated
Connecting AI agents
MCP in Multogram: how to connect Claude Desktop, Claude Code, Cursor, VS Code, Codex and other AI clients to the application. Where to get the mcp.exe MCP server, downloading the archive, setting up the client and checking the connection.
If the agent won't connect
Troubleshooting MCP in Multogram: the agent doesn't see the tools, "Multogram is not running", versions don't match, no connection to the server, no device slot, AI session limit, refusals and errors — what each message means and what to do.