> For the complete documentation index, see [llms.txt](https://docs.meetalto.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.meetalto.ai/mcp/recipes.md).

# Workflows and prompts

Use example prompts and step-by-step workflows to get useful work done with your assistant.

Connect with the [quickstart](https://docs.meetalto.ai/mcp/quickstart) first. These example requests depend on the tools available to your connection. The assistant should inspect those tools and explain missing permissions.

## Start with a complete workflow

* [Update knowledge](https://docs.meetalto.ai/mcp/workflows/knowledge): find a gap, review an answer, save it and check retrieval.
* [Prepare and test an agent](https://docs.meetalto.ai/mcp/workflows/agent): review a playbook, inspect save results and test before deciding to enable it.
* Prepare a campaign: draft, validate and review its audience before activation.

For better results, give the assistant a goal, the exact account or resource, relevant dates and the changes you authorize. Tell it when you want a proposal first. Conversation instructions do not reduce granted permissions; review your client's tool approval settings too.

## Understand your workspace

> Tell me which Meet Alto workspace this connection uses. Read its setup progress and explain the next unfinished step. Do not change any settings.

Check the returned workspace name before continuing.

**Check the result:** setup completion, saved profile details and account connection health are different things. The assistant should identify missing information without guessing.

## Find an account that needs attention

> List connected accounts in my workspace. Separate activity state from authentication health and explain any reported issues in plain language. Recommend the next step for each account; do not reconnect or disconnect anything.

An inactive account and an account that requires authorization again are not interchangeable. Follow the [account reconnection guide](https://docs.meetalto.ai/using-meet-alto/accounts/reconnect) when needed.

## Investigate a workflow

> List the campaigns I can access. Help me identify the campaign for my lead guide and explain its saved configuration and status. Do not publish, enable, or send anything.

Give the exact campaign name or ID when you have it. Similar names are a reason to clarify the target.

## Prepare a campaign

You can start simply:

> Create a comment-to-DM campaign for my free guide.

The assistant should resolve the account and supported trigger, ask for essential missing facts (such as which guide and its real URL), and propose a short conversation. It should state its choices instead of expecting you to know configuration fields. Your request can authorize saving the draft; you do not need a separate approval round for every step. A request for a proposal only should remain a proposal.

For example, a useful guide flow might greet the commenter, offer a “Send guide” button, and deliver the guide after the interaction. Email capture, follow requirements, branch choices, integrations, reminders and an AI conversation are optional: include only what helps your goal and works on the selected account. Do not invent a resource URL or enable every action. Inspect connection capabilities and use the app for unavailable setup, media or integration steps.

For new images or videos, open the campaign editor in the app and upload them in the appropriate message block. For an integration action, first connect and configure the destination in [Settings → Integrations](https://docs.meetalto.ai/using-meet-alto/integrations), then select it in the campaign editor. Media upload and integration destination discovery are outside this documented release. Reuse only verified, authorized references; never invent storage keys or destination IDs.

For more control, provide the account, desired trigger, content to send, audience and intended outcome. Use verified business details.

> Read the selected account and available campaign options. Propose a campaign for the goal I provide. Show the trigger, opening message, audience, next action and any missing prerequisites. Do not save or activate it yet.

When saving is authorized, have the assistant use `create_campaign`, run `validate_campaign`, and read it with `get_campaign`. Ask it to explain the saved trigger, opening, delivery, optional stages, assumptions and remaining blockers. New campaigns default to draft. When creating a campaign, you can omit `audience` to inherit the account's current AI reply rules. Creating a draft does not change those rules. If you explicitly set AI reply permission, supply the current audience revision.

Validation checks whether a saved campaign can activate. Successful saves, reads and validation can return advisory `warnings` with a field `path` and an action to take. Advisory warnings identify starter copy to review, another live campaign on the same account that answers the same comments first (so one of them never sends), or another campaign on the account with the same name; they do not require optional features or judge all possible promises. Generic starter copy can be executable without fulfilling your goal; warnings do not block activation. No warning is a guarantee of quality or proof that a promised link is correct. Verify the actual resource, links and saved enabled stages before activation.

**Check:** the saved account, trigger, message, audience and draft status match what you approved. For edits, follow the campaign patch and clear rules.

> Change only the opening message of my guide campaign to “Ready for your guide?”

The assistant should read the current campaign, edit only the appropriate opening fields, validate and read back. An A/B opening uses its variants; it must not reset delivery, collection or other stages to defaults.

If input is rejected, use the returned field paths and `get_operation_schema` to correct it. Missing permissions require the offered access flow; stale revisions require a fresh read and reconsidering the edit. An uncertain write is different: follow [request recovery](https://docs.meetalto.ai/api/requests) before attempting another write.

Activation is a separate decision. When the exact saved campaign is ready and you want it live, approve activation. The assistant uses `activate_campaign` with current configuration and audience revisions and then reads the result. Activation requires `campaigns:publish` and can start the normal workflow for eligible events. A validation result alone does not activate it. Activation also requires a usable DM keyword or the applicable specific post/story selection; story mentions do not require keywords.

Use the [campaign guide](https://docs.meetalto.ai/using-meet-alto/campaigns) to understand the product choices. The same review pattern applies to sequences: draft, read, validate and separately publish. Edits to active workflows can require publishing access.

### Campaign patch and clear rules

Read the campaign first and pass its current configuration/audience revisions. For a narrow edit, use `update_campaign` with `config_patch`, for example `{"openingMessage":{"text":"Your new greeting"}}`. Omitted fields stay unchanged. The supported message/settings objects merge their named fields; opening A/B settings also merge named fields. Arrays, branches, variants and scripts replace as whole units. Use empty arrays to clear a collection, empty strings to clear text, and `enabled: false` to disable a stage. In the shared follow-up message, `script: null` or `imageKey: null` removes that optional value.

When opening A/B testing is enabled, edit `openingMessage.abTest.variants` to change the greeting, button text or branches. The ordinary opening fields mirror variant A; a patch that conflicts with that variant is rejected. You can disable A/B and edit the ordinary opening in the same patch. An explicitly supplied `winnerVariantId` must match a variant in the resulting enabled A/B test.

`config` still replaces the authored configuration. Do not send it together with `config_patch`. Server-owned binding fields are not writable. Live edits require publishing permission, even if you did not ask to change the campaign's status. Drafts may remain incomplete, but malformed A/B structures must be corrected.

## Improve knowledge

> Review the knowledge sources and gaps I can access. Identify questions that need an answer and propose concise text snippets for my review. Keep proposals separate from saved knowledge.

Supply verified business facts before approving a snippet. Do not let the assistant invent prices, eligibility, delivery dates, or policies.

For a document, use `get_knowledge_upload_options`, create the upload, transfer the exact bytes, then complete it. Attaching a file in chat does not guarantee that the host can transfer those bytes. If it cannot, upload through [Knowledge Sources in the app](https://app.meetalto.ai/settings?tab=knowledge), then ask the assistant to read the source's status.

Keep the completion's original key and input. A lost completion response must not cause a new ingestion; follow the [upload recovery rules](https://docs.meetalto.ai/api/requests#recover-a-knowledge-upload). Read the returned source/job references, wait for processing, and test retrieval. The assistant should say “queued” until processing succeeds. Persistent ingestion failure requires inspection/support; `retry_knowledge_cleanup` is for deletion cleanup only.

## Review customer questions

> Read a bounded sample of conversations for the account and date range I specify. Group recurring questions and identify where answers or handoffs could improve. State the sample size and filters. Summarize without copying customer contact details, and do not send or modify anything.

A sample can reveal useful questions without representing every conversation. Ask the assistant to state that limit. Reading conversations does not mark them read.

## Organize a contact list

> Find contacts matching the criteria I give you and show a proposed list name and membership count. Explain the filters and whether you reviewed all matching pages. Do not create a list, change membership or export data until I approve the selection.

Approve an exact selection before a write or export. Contact exports are bounded pages, so one successful export is not automatically the entire database. Contact identifiers used for reading and list selection can differ; let the assistant use the actual returned IDs and current schema.

For individual membership changes, use `list_ids_add` or `list_ids_remove` in `update_contact`. Read the current membership revision before a full replacement; see [contact editing rules](https://docs.meetalto.ai/using-meet-alto/contacts).

## Find your biggest contacts

> For each of my Instagram accounts, find contacts with more than 10,000 followers who follow us. Show the first few and the total. Say how many contacts' follower counts aren't known yet.

Meet Alto knows a contact's follower count, and whether they follow you, only after they DM that Instagram account, and it uses counts checked in the last day. Contacts without a current count are left out of the results, not counted as a no. To look them up, add the same **Followers** or **Follows you** filter in [Contacts](https://docs.meetalto.ai/using-meet-alto/contacts#find-and-export-contacts) and select the update link under the filters.

## Analyze what customers write

> Read an even sample of what customers wrote in comments and DMs over the last 30 days. Group the main topics and how people feel, with how often each comes up. Say how many messages you read of how many.

`sample_inbox_text` returns up to 2,000 messages spread across the period, in pages of about 200, with the total for the period. It returns text only, without names, and uses no Meet Alto AI credits: your AI app does the analysis. Ask for a shorter period or particular accounts to focus it.

## Save a segment or an alert

> Save a segment called "Big followers" for contacts with more than 10,000 followers who follow our Instagram.

> Save a Slack alert for when a customer waits more than 30 minutes for a reply. Leave it off.

A segment keeps itself up to date as contacts match or stop matching. An alert is saved switched off: a workspace owner turns it on in **Settings → Alerts** in Meet Alto. Alerts send to the Slack channel connected in Meet Alto, and mention and waiting-customer alerts work for Instagram and Facebook accounts connected through Facebook. See [Get alerts](https://docs.meetalto.ai/using-meet-alto/integrations/slack#get-alerts).

## Review results

> Read the available analytics for this account and date range. Explain the metric definitions and any missing data before suggesting one improvement.

Use explicit dates and account names. Delivered messages, clicks, and captured contacts are different outcomes.

**Check:** the assistant states the period, filters, data limitations and evidence behind its recommendation. Ask it to separate agent session outcomes from engagement events, and lifetime variant statistics from metrics that use your selected date range.

## Preview a reply without sending

> For the sample customer question I provide, preview a reply using saved brand guidance and knowledge. Explain any information the answer lacks. Do not send it, save a draft or change settings.

Sample previews differ from generating a response to an existing inbox item. A successful inbox reply generation consumes an AI credit. [Understand generation and costs](https://docs.meetalto.ai/mcp/actions#understand-costs-and-limits).

## Move from a proposal to a change

Ask for the exact resource, requested changes, permissions, and expected effects first. Approve a specific change once you understand it. For an uncertain result, inspect the current state before retrying; [retries have special rules](https://docs.meetalto.ai/api/requests).

Direct inbox sending and moderation, billing changes, team administration, and broader integration management are outside this documented MCP release. Use the app for supported dashboard workflows.