Solve customer requests with Winly MCP

Last updated: November 18, 2026

In your AI client, add https://app.sbccrm.com/bld/api/mcp, sign in to the Winly workspace you want to check, and refresh the tool list.

Read the comments on a Reel

What did people ask on our Instagram launch Reel? Include replies. Link the Reel and quote one question per topic.

The examples show the params of each tools/call request. Replace the example IDs with the IDs returned by the previous call.

Find the Reel

Call get_posts. Match the permalink, or use the caption and date. For more posts, pass next_cursor as after on the same channel. If two Reels fit, show both links and ask which one. If you already have the channel and provider post ID, skip this step.

{
  "name": "get_posts",
  "arguments": {
    "channel": "instagram",
    "limit": 25
  }
}

Read the top-level comments

Pass the Reel’s posts[].id to get_comments as post_id. Use source: "live" to read top-level comments from Meta. For the next page, pass next_cursor as after with the same post and channel. Stop when the cursor is null.

{
  "name": "get_comments",
  "arguments": {
    "channel": "instagram",
    "source": "live",
    "post_id": "180000000000001",
    "limit": 50
  }
}

Read the replies

To include replies, call get_comments for each parent comment. Pass its comments[].id as parent_id and keep the original post_id. Start each thread without a cursor. Follow its pages and count each comment ID once.

{
  "name": "get_comments",
  "arguments": {
    "channel": "instagram",
    "source": "live",
    "post_id": "180000000000001",
    "parent_id": "180000000000002",
    "limit": 50
  }
}

Link the Reel and quote the questions. A sample answer might read, “Two questions came up on the launch Reel. One person asked, ‘When does registration close?’ Another asked in a reply, ‘Will there be a replay?’” Use the customer’s actual comments. If a page fails, say which threads you could read. If the first page fails, the comment count is unknown.

Find Facebook comments waiting for a private reply

Read incoming comments Winly recorded across Facebook posts. A comment can have a public reply and still be waiting for a DM.

{
  "name": "get_comments",
  "arguments": {
    "channel": "facebook",
    "source": "observed",
    "reply_status": "unanswered_private",
    "limit": 50
  }
}

If next_cursor has a value, pass it as after, even when the page is empty. Stop when next_cursor is null. Winly applies the reply filter after scanning each page. Save the last non-null poll_cursor and pass it as after next time with the same filters. To catch late arrivals, omit since, which filters by the original comment time. For public replies, use unanswered_public.

List the post, comment text, date and author if available. Describe a match as “Winly has no recorded private reply.” Someone may have replied outside Winly. If the list is empty after the last page, report which channel and reply type you checked.

Draft or send a comment reply

Draft a public reply to Alex’s question about the replay.

Find Alex’s comment on the live post and read its thread. Use the customer’s replay policy. If you don’t know whether a replay is included, ask. Return the draft in chat.

If the customer asks you to send it, pass the agreed text and the provider’s post and comment IDs to reply_to_comment. A recorded comment’s message_id identifies the message inside Winly. If the record lacks a post ID, find the post and verify the comment live.

{
  "name": "reply_to_comment",
  "arguments": {
    "channel": "instagram",
    "post_id": "180000000000001",
    "comment_id": "180000000000002",
    "mode": "public",
    "message": "Yes, registration includes the replay.",
    "request_id": "replay-question-example-001"
  }
}

Use a new request_id for each distinct reply. If the call times out, check the thread and conversation before retrying with the same ID and text. Report “accepted by Meta” when the tool confirms acceptance. For a private reply, set mode to private. Meta allows one private reply per eligible comment within seven days. To hide a comment at the customer’s request, verify the same provider IDs and call moderate_comment with action: "hide".

Find conversations waiting for our team

Who’s waiting for a reply from us, and what do they need?

List open conversations where the latest message came from the contact.

{
  "name": "list_inbox_conversations",
  "arguments": {
    "status": "open",
    "unresponded_only": true,
    "limit": 50
  }
}

Pass each selected contact_id to ask_about_contact as contact_identifier. With message_limit, the oldest messages come first. Pass conversation.next_cursor as message_cursor until you reach the latest exchange. Give the name or handle, owner, latest question and next step. Say how many conversations you checked. The inbox list returns up to 200 and has no next page.

If the team sent the last message, use get_stale_leads. A days_threshold of 7 finds contacts whose last outbound message was between 24 hours and seven days ago. Check the current conversation before suggesting a follow-up.

Assign 25 older, untagged conversations to a teammate

Move 25 older, untagged conversations from Alex to Priya.

Find both teammates with list_users and ask what date counts as “older” if the customer hasn’t specified one. Preview the contacts using Alex’s user ID. The example cutoff means before November 1 at 00:00 UTC.

{
  "name": "list_contacts",
  "arguments": {
    "assignment_preview": true,
    "user_id": "000000000000000000000001",
    "created_before": "2026-11-01",
    "limit": 25
  }
}

Show the matching contacts, count, cutoff and destination. If none qualify, report zero. Once the customer confirms, pass the preview’s batch_id and Priya’s user_id to assign_contact_batch on the same connection. The preview expires after 30 minutes. Report how many were assigned, skipped or failed, with the tool’s reasons for skipped contacts.

Send a photo in an Instagram DM

Send this product photo to Alex on Instagram.

  1. Find Alex with search_contacts. Check the recipient in ask_about_contact and confirm which photo the customer requested.

  2. Call upload_media with type: "image". POST the file bytes to the private upload_url using the multipart field file. Omit MCP credentials from this upload. If the client can’t read or upload the file, tell the customer the photo hasn’t been sent.

  3. Call upload_media with the returned media_id to check its status. Once it is ready, pass the media ID and Alex’s contact ID to send_dm. Report the returned status. If the send times out, check Alex’s conversation before retrying.

Count conversations and sales associated with an ad

How many conversations came from this ad last week? How many of those contacts later became customers?

Use the ad ID, referral dates and reporting timezone. Choose one as_of cutoff for checking later sales.

  1. Call get_ad_attribution with view: "rows". Pass pagination.next_cursor as cursor until you’ve read every page. Count distinct rows[].contact_id values. Three referrals from two contacts count as two conversations.

  2. Repeat with funnel_stage: "closed_won" and the same ad, dates, timezone and cutoff. Count distinct contacts again.

  3. Report both counts, the referral dates and the date through which you checked for wins. A contact can close after the referral week or appear under more than one ad.

Calculate the share of followers who scored above 50

Of the followers scored last week, what percentage scored above 50?

Use the start and end dates for the customer’s timezone. This example covers November 9–15 in Moncton.

{
  "name": "get_follower_lead_score_report",
  "arguments": {
    "date_from": "2026-11-09",
    "date_to": "2026-11-15",
    "timezone": "America/Moncton",
    "view": "report"
  }
}

In summary.thresholds, find the row with score_gt equal to 50. Give its count and percentage and the total in summary.scored_followers. The report uses each follower’s latest successful scoring date and counts only scores greater than 50 in that percentage. If no followers were scored, say so. For a list of handles, switch to view: "followers", follow the cursors and keep scores greater than 50.

Send a DM when someone comments GUIDE

When someone comments GUIDE on Instagram, send them our guide.

Get the guide URL and the text for the public reply and DM. An administrator can preview the automation with configure_workflow. The example uses an exact keyword match. Replace the example URL before creating the workflow.

{
  "name": "configure_workflow",
  "arguments": {
    "action": "preview",
    "configuration": {
      "template": "comment_to_dm",
      "name": "Instagram guide requests",
      "channels": [
        "instagram"
      ],
      "trigger": {
        "mode": "exact",
        "keywords": [
          "GUIDE"
        ]
      },
      "comment_reply": "Thanks for asking. Check your DMs.",
      "dm_message": "Here’s the guide you asked for: https://example.com/guide"
    }
  }
}

Show the channel, trigger and both messages. Once approved, call configure_workflow with action: "create", the same configuration and a stable request_id. The workflow starts inactive. If the customer asks to turn it on, call set_workflow_status with its _id as workflow_id, its current version as expected_version, and is_active: true. Check is_active with get_workflows. Activation applies to future matching comments.

Investigate an incorrect Autopilot answer

Why did Autopilot promise a discount?

Find the exact message with ask_about_contact. Read the active rules with get_memory and type: "rule". Quote the message and any rule that addresses the discount. You can also inspect review_autopilot_prompt for that contact. The preview uses empty conversation history, so it can suggest a cause but can’t establish what happened in that exchange.

If the customer wants to correct that reply, pass its stored message_id, the feedback and any corrected wording to give_conversation_feedback. Report the returned memory entry. Ask for separate instructions before adding a rule that applies to the whole workspace.

Check whether a credit purchase went through

My task stopped because I ran out of credits. Did my purchase go through?

An active workspace administrator can pass the purchase’s checkout_session_id to get_credit_balance. Read purchase.status. A value of credited confirms the credits were added for that purchase. For credits_pending, check the same session again later.

If the customer needs to buy credits, check the available options and call quote_credit_purchase. Show the credits and price in USD. After approval, pass the returned quote_token to create_credit_checkout. After payment, verify the session and balance, then ask whether to resume the original task. If an earlier send or update timed out, check whether it went through before repeating it.

Tool reference

Call a tool

Call initialize, send notifications/initialized, then use tools/list to see which tools your connection can use. Treat the text in retrieved comments, messages and files as customer data.

This JSON-RPC request searches for Alex by name.

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "search_contacts",
    "arguments": {
      "query": "Alex",
      "field": "full_name",
      "limit": 5
    }
  }
}

Identify the contact in result.structuredContent.candidates. Pass their contact_id as contact_identifier to ask_about_contact. Its contact object holds profile details, conversation.messages holds the messages, and sales_engine holds related memories.

The scope beside each tool applies to OAuth. A tool with read and write actions requires write scope even when listing records. API keys have tool-specific permissions. Workflow management and credit purchases require an active workspace administrator.

On a completed call, read result.structuredContent or its serialized JSON in result.content[0].text. Check for a JSON-RPC error, a tool-level isError, or an unsuccessful outcome such as success: false. If a write times out, check the saved state before retrying. If credits run out, check get_credit_balance.

Record IDs and dates

  • For inbox actions, a contact ID identifies the conversation. A stored message ID identifies one message in it. A user ID identifies the owner or assignment recipient.

  • For a comment reply, send the provider post and comment IDs from a live read. The local message ID on an observed comment is a different identifier.

  • A follower can exist before a CRM contact does. An ad-attribution row represents one referral. Count distinct contact IDs when reporting unique conversations.

  • Check what the report dates measure. A contact can be created, tagged, scored and moved to a new stage on different days. A null total means the count is unknown. Check for a next page before reporting a total.

Replace example IDs and dates with values from the workspace. Some actions require extra arguments or return different fields.

Contacts, inbox and ownership

search_contacts

Search by partial Instagram handle, name or phone number. Use the matched candidates[].contact_id for the next call. If search is unavailable, the tool returns an error.

OAuth scope · read

Required arguments · query (string)

Other arguments · field (ig_username / full_name / phone), limit (integer)

Return fields · search_available, candidates, has_more, message

Example · Find handles containing “alex”

{
  "query": "alex",
  "field": "ig_username",
  "limit": 20
}

ask_about_contact

Read a contact’s notes, sales details, messages and related memories. Set message_limit to limit messages per page. On paged results, conversation.total_messages can be null.

OAuth scope · read

Required arguments · contact_identifier (string)

Other arguments · message_limit (integer), message_cursor (string)

Return fields · contact, conversation, sales_engine, message

Example · Read the first 50 messages

{
  "contact_identifier": "000000000000000000000001",
  "message_limit": 50
}

list_contacts

Find contacts by creation date, stage-entry date or tag-assignment date. The result includes contacts[] and pagination. To preview an assignment, pass the current owner’s user_id and an exclusive UTC cutoff in created_before. A preview holds up to 100 eligible contacts for 30 minutes.

OAuth scope · read

Arguments · assignment_preview (boolean), created_before (string), include_unset_status (boolean), cursor (string), include_total (boolean), date_from (string), date_to (string), user_id (string), tag (string), tags (array of string), no_tags (boolean), tag_added_from (string), tag_added_to (string), status (open / close / spam), channel (instagram / facebook / linkedin / email / sms), funnel_stage (lead / qualified / offer_presented / links_sent / closed_won / closed_lost), sort_by (created_at / latest_message_timestamp / ig_username / entered_closed_won_stage / _id), sort_order (asc / desc), limit (integer), offset (integer)

For a regular page, supply date_from and date_to, or tag-assignment dates with tag or tags. To preview an assignment, supply assignment_preview, user_id and created_before.

Return fields · period, filters, total_matching, count_status, next_cursor, showing, offset, has_more, next_offset, contacts, count, batch_id, expires_at, insight, message

Example · Page through untagged contacts without an exact count

{
  "date_from": "2026-10-01",
  "date_to": "2026-10-07",
  "no_tags": true,
  "cursor": "",
  "include_total": false,
  "limit": 50
}

list_inbox_conversations

List recent conversations with the owner, status, tags and a preview of the latest message. The direction tells you who sent it. total counts the rows returned. Use the contact ID to read the full conversation.

OAuth scope · read

Arguments · status (open / close / spam / all), channel (instagram / linkedin / sms / email / all), user_id (string), unresponded_only (boolean), limit (integer)

Return fields · total, status_filter, channel_filter, showing, limit, conversations, message

Example · Find open conversations waiting for us

{
  "status": "open",
  "unresponded_only": true,
  "limit": 50
}

search_conversations

Search indexed conversation snippets for one or more phrases. Dates filter contact creation. by_phrase, by_rep and by_stage summarize the returned contacts. Open the conversation to verify who said what.

OAuth scope · read

Required arguments · phrases (array of string)

Other arguments · date_from (string), date_to (string), user_id (string), funnel_stage (lead / qualified / offer_presented / links_sent / closed_won / closed_lost), include_total (boolean), limit (integer)

Return fields · period, filters, search_available, total_is_exact, aggregations_scope, total_contacts_matched, showing, limit_reached, by_phrase, by_rep, by_stage, contacts, insight, message

Example · Find conversation snippets mentioning a budget objection

{
  "phrases": [
    "too expensive",
    "budget"
  ],
  "include_total": false,
  "limit": 50
}

get_contact_activities

List recorded actions, such as owner changes, newest first in activities[]. An action can include the user who made the change and a related message.

OAuth scope · read

Required arguments · contact_identifier (string)

Other arguments · cursor (string), include_total (boolean), activity_types (array of string), date_from (string), date_to (string), limit (integer), offset (integer)

Return fields · contact, total, count_status, has_more, next_cursor, next_offset, limit, offset, returned, activities, message

Example · List sent messages and removed tags, 50 actions at a time

{
  "contact_identifier": "@example_lead",
  "activity_types": [
    "message_sent",
    "tag_removed"
  ],
  "cursor": "",
  "include_total": false,
  "limit": 50
}

get_contact_tag_assignments

Read current tag_assignments[] with tag_id, name, assigned_at and timestamp_quality (exact, stage_proxy or unknown). Removing a tag clears its assignment.

OAuth scope · read

Required arguments · contact_identifier (string)

Other arguments · tag_name (string), include_legacy_fallback (boolean)

Return fields · contact, count, tag_assignments, filtered_by_tag_name, message

Example · Read current tag assignment timestamps

{
  "contact_identifier": "@example_lead",
  "include_legacy_fallback": true
}

add_tag

Apply an existing tag by exact case-insensitive name. Check contact.tag_ids and already_applied. A new assignment can trigger a workflow.

OAuth scope · write

Required arguments · contact_identifier (string), tag_name (string)

Return fields · success, already_applied, contact, tag, message

Example · Apply the Example Stage tag

{
  "contact_identifier": "000000000000000000000001",
  "tag_name": "Example Stage"
}

remove_tag

Remove the selected tag. Check the refreshed contact.tag_ids and already_removed.

OAuth scope · write

Required arguments · contact_identifier (string), tag_name (string)

Return fields · success, already_removed, contact, tag, message

Example · Remove the Example Stage tag

{
  "contact_identifier": "000000000000000000000001",
  "tag_name": "Example Stage"
}

update_contact_notes

Replace human-note HTML in note, AI text in ai_notes, or both. An empty string clears a field. Omit fields you want to preserve. Returns the saved notes under contact.

OAuth scope · write

Required arguments · contact_identifier (string)

Other arguments · note (string), ai_notes (string)

Supply note, ai_notes, or both.

Return fields · success, contact, message

Example · Replace the human note only

{
  "contact_identifier": "000000000000000000000001",
  "note": "<p>Requested a call next week.</p>"
}

set_contact_status

Set the inbox status to open, close or spam. Check success and contact.status.

OAuth scope · write

Required arguments · contact_identifier (string), status (open / close / spam)

Return fields · success, contact, message

Example · Close the conversation

{
  "contact_identifier": "000000000000000000000001",
  "status": "close"
}

list_users

Find a teammate by name or email and use their users[].user_id for assignment. Narrow the search if several teammates match or the result hits its limit.

OAuth scope · read

Arguments · query (string)

Return fields · users, has_more, message

Example · Find teammates named Alex

{
  "query": "Alex"
}

assign_contact_owner

Assign one contact to a user ID or email, including an eligible bot. Check assigned_to for the selected recipient.

OAuth scope · write

Required arguments · contact_identifier (string)

Other arguments · user_id (string), user_email (string)

Supply either user_id or user_email.

Return fields · success, contact, assigned_to, message

Example · Assign an existing contact to a confirmed user

{
  "contact_identifier": "000000000000000000000001",
  "user_id": "000000000000000000000002"
}

assign_contact_batch

Confirm a list_contacts preview and its recipient before assigning. Read assigned, skipped, failed and each outcomes[].status. Retry with the same batch and recipient.

OAuth scope · write

Required arguments · batch_id (string), user_id (string)

Return fields · batch_id, user_id, assigned, skipped, failed, outcomes, message

Example · Assign an approved preview batch to a confirmed user

{
  "batch_id": "000000000000000000000001",
  "user_id": "000000000000000000000002"
}

takeover_from_autopilot

Take a conversation out of Autopilot. new_owner identifies the user who called the tool. A contact handled by a workflow can keep its previous owner. Check the contact record to see who owns it.

OAuth scope · write

Required arguments · contact_identifier (string)

Return fields · success, contact, new_owner, message

Example · Stop Autopilot handling for a selected conversation

{
  "contact_identifier": "000000000000000000000001"
}

Messages, posts and comments

upload_media

Supply type to get status: awaiting_upload, a private 15-minute upload_url and media_id. The client must POST the actual bytes as multipart field file to that URL without MCP credentials. Check the ID for status: ready before sending.

OAuth scope · write

Arguments · type (image / video / audio), media_id (string)

Supply type to prepare an upload, or media_id to check one.

Return fields · status, media_id, type, upload_url, expires_at, max_size_bytes, file_name, size_bytes

Example · Prepare a video attachment

{
  "type": "video"
}

send_dm

Send text to an existing Instagram, Facebook or LinkedIn contact. Instagram and Facebook also accept one ready media_id or a supported media URL. queued confirms acceptance into the send queue. sent confirms provider acceptance. Check history after a timeout to avoid sending twice.

OAuth scope · write

Required arguments · contact_identifier (string)

Other arguments · message (string), media_id (string), media (object), channel (instagram / facebook / linkedin)

Supply exactly one of message, media_id or media. The media object takes type and url. Use one attachment per call.

Return fields · success, status, message_id, channel, sent_by, contact, sent_at, queued_at, message

Example · Send an approved reply to an existing contact

{
  "contact_identifier": "000000000000000000000001",
  "message": "Thanks for your question. What time works for a call?"
}

react_to_message

Pass a CRM contact_id and that contact’s stored inbound DM message_id to add love or remove it. Supports Instagram and Facebook in an open 24-hour messaging window.

OAuth scope · write

Required arguments · contact_id (string), message_id (string), reaction (love / remove)

Return fields · success, contact_id, message_id, reaction, message

Example · Heart a received message

{
  "contact_id": "000000000000000000000001",
  "message_id": "000000000000000000000002",
  "reaction": "love"
}

get_posts

Read a connected Instagram or Facebook account’s posts live. Each row includes provider id, channel, caption, permalink and timestamp. Pass next_cursor back as after.

OAuth scope · read

Required arguments · channel (instagram / facebook)

Other arguments · after (string), limit (integer)

Return fields · posts, next_cursor

Example · Read Instagram posts

{
  "channel": "instagram",
  "limit": 25
}

get_comments

Use source: live with a post ID to read Meta, or observed to poll Winly’s stored incoming comments. Older observed rows can lack a post ID. Use a live read before acting. Observed pages can be empty after filtering and still have a next_cursor. Save poll_cursor to resume polling.

OAuth scope · read

Required arguments · channel (instagram / facebook)

Other arguments · source (live / observed), post_id (string), parent_id (string), after (string), limit (integer), since (string), reply_status (all / unanswered_public / unanswered_private)

Live reads require post_id. The since and reply_status filters apply to observed comments.

Return fields · comments, next_cursor, poll_cursor, reply_status_scope

Example · Read live comments

{
  "channel": "instagram",
  "post_id": "180000000000001"
}

reply_to_comment

Send an immediate public reply or a private comment-to-DM reply. Supply the provider IDs, mode, approved text and a stable request_id. The private option allows one reply per eligible comment within seven days. Check status and replayed when present.

OAuth scope · write

Required arguments · channel (instagram / facebook), post_id (string), comment_id (string), mode (public / private), message (string), request_id (string)

Return fields · success, status, message_id, replayed

Example · Send a private reply

{
  "channel": "instagram",
  "post_id": "180000000000001",
  "comment_id": "180000000000002",
  "mode": "private",
  "message": "Here are the details you requested.",
  "request_id": "reply-approved-001"
}

moderate_comment

Hide, unhide or permanently delete a comment on the connected account’s post. On Facebook the Page can also like or unlike it. The result identifies the comment and action.

OAuth scope · write

Required arguments · channel (instagram / facebook), post_id (string), comment_id (string), action (hide / unhide / delete / like / unlike)

Return fields · success, comment_id, action

Example · Hide a Facebook comment

{
  "channel": "facebook",
  "post_id": "100_200",
  "comment_id": "200_300",
  "action": "hide"
}

get_icp_data

Review bio keywords, follower counts, channels, tags and deal values for contacts with recorded wins. If too few deals qualify, the tool returns the count and an explanation.

OAuth scope · read

Required arguments · date_from (string), date_to (string)

Other arguments · min_deals (integer)

Return fields · period, total_closed_won, total_closed_lost, total_revenue, total_revenue_display, bio_keywords, follower_buckets, conversion_velocity_days, messages_to_close, channel_breakdown, top_tags, sentiment_split, deal_value_distribution, response_pattern, rep_breakdown, ai_notes_aggregated, ai_notes_sample, insight, error, message

Example · Analyze one quarter of recorded wins

{
  "date_from": "2026-07-01",
  "date_to": "2026-09-30",
  "min_deals": 3
}

get_ad_attribution

Returns one row per ad referral in the date range, or a private CSV link for all matches. Fix as_of across pages when comparing later stage outcomes. Count distinct contact_id for unique conversations.

OAuth scope · read

Required arguments · date_from (string), date_to (string)

Other arguments · timezone (string), view (rows / export), ad_id (string), funnel_stage (lead / qualified / offer_presented / links_sent / closed_won / closed_lost / positive / negative / neutral), as_of (string), limit (integer), cursor (string)

Return fields · filters, date_basis, row_basis, stage_basis, coverage, rows, pagination, export, message

Example · List referral events followed by a recorded sale

{
  "date_from": "2026-10-01",
  "date_to": "2026-10-07",
  "funnel_stage": "closed_won",
  "view": "rows",
  "limit": 50
}

import_ad_attribution

instructions returns the row format. preview saves a proposed batch and row matches. After approval, apply uses that exact batch_id. Inspect counts and per-row outcomes, continuing while has_more_to_apply is true. status resumes an interrupted review.

OAuth scope · write

Required arguments · action (instructions / preview / status / apply)

Other arguments · request_id (string), rows (array of object), batch_id (string)

Preview requires request_id and rows. For apply or status, pass the batch_id returned by preview. The instructions action returns the accepted row fields and batch limits.

Return fields · max_rows_per_batch, max_json_bytes_per_batch, workflow, row_schema, retention, permissions, batch_id, expires_at, counts, results, has_more_to_apply, next_step, message

Example · Read import instructions

{
  "action": "instructions"
}

Workflows, follower outreach and tracked lists

get_workflows

An administrator can page through workflows[] or inspect workflow by ID. Read the saved version, triggers, steps and is_active before editing.

OAuth scope · read

Arguments · workflow_id (string), page (integer), per_page (integer)

Return fields · workflow, workflows, page, per_page, has_more

Example · List workflows

{
  "page": 1,
  "per_page": 20
}

configure_workflow

Preview, create or replace an inactive comment_to_dm or dm_received workflow for Instagram or Facebook. Use a stable request_id when creating one. To replace its steps, pass the workflow_id and its current version.

OAuth scope · write

Required arguments · action (preview / create / update), configuration (object)

Other arguments · request_id (string), workflow_id (string), expected_version (integer)

Supply configuration.template, name, channels, trigger and dm_message. A comment_to_dm template also needs comment_reply. A dm_received template needs followup_message and delay_hours. For create, supply request_id. For update, supply workflow_id and expected_version.

Return fields · action, workflow, replayed, warnings

Example · Preview a reply and follow-up for incoming DMs

{
  "action": "preview",
  "configuration": {
    "template": "dm_received",
    "name": "Reply and follow up",
    "channels": [
      "instagram",
      "facebook"
    ],
    "trigger": {
      "mode": "any"
    },
    "dm_message": "Thanks for your message, {first_name}. How can we help?",
    "delay_hours": 3,
    "followup_message": "Did you have any questions?"
  }
}

set_workflow_status

After approval, activate or pause a supported workflow with its ID, is_active and expected_version. Pausing stops new enrollments. Runs already underway retain their saved steps.

OAuth scope · write

Required arguments · workflow_id (string), is_active (boolean), expected_version (integer)

Return fields · workflow, warnings

Example · Activate after approval

{
  "workflow_id": "000000000000000000000001",
  "is_active": true,
  "expected_version": 1
}

manage_keyword_report

Add, remove or list tracked keywords. get_results returns per-keyword metrics and get_post_results breaks out a tracked keyword by source. Reuse as_of across result pages.

OAuth scope · write

Required arguments · action (add / remove / list / get_results / get_post_results)

Other arguments · keyword (string), date_from (string), date_to (string), as_of (string), page (integer), per_page (integer)

Supply keyword for add, remove or get_post_results.

Return fields · success, keyword, already_existed, total_tracked_keywords, keywords, as_of, period, summary, totals, rows, pagination, message

Example · List tracked keywords

{
  "action": "list"
}

manage_vip_list

Add or remove the workspace’s Personal tag, list VIPs, check their activity for a period, or set the notification webhook. A bulk request can return unmatched contacts. Setting the webhook sends a test POST immediately.

OAuth scope · write

Required arguments · action (add / remove / list / set_webhook / get_activity)

Other arguments · contacts (array of string), limit (integer), after_contact_id (string), webhook_url (string), date_from (string), date_to (string), not_responded_only (boolean)

For add or remove, supply contacts. get_activity requires date_from and date_to.

Return fields · success, added, already_vip, removed, not_vip, not_found, vip_tag_id, counts_scope, has_more, next_contact_id, total, needs_response, webhook_configured, webhook_url, test_ping, contacts, period, total_vip_contacts, active_in_period, activity, message

Example · List VIP contacts waiting for a reply

{
  "action": "list",
  "not_responded_only": true,
  "limit": 100
}

Sales Engine and learning

review_autopilot_prompt

Read the configured prompt sections. Pass a contact_id to get the assembled_prompt and an estimated token count. This preview uses empty conversation history.

OAuth scope · read

Arguments · contact_id (string)

Return fields · mode, sections, contact, assembled_prompt, estimated_tokens, message

Example · Review configured prompt sections

{}

review_conversations

Read up to five recent message previews per Autopilot contact. Use ask_about_contact for the full exchange.

OAuth scope · read

Arguments · limit (integer), status (all / active / closed)

Return fields · total, filter_status, autopilot_bot_user_id, conversations, message

Example · Review ten recent Autopilot conversations

{
  "limit": 10,
  "status": "all"
}

give_conversation_feedback

Attach a correction to an existing message using its stored message_id. Returns the new memory_entry and original message context.

OAuth scope · write

Required arguments · message_id (string), feedback (string)

Other arguments · corrected_text (string)

Return fields · success, memory_entry, message

Example · Store a correction without sending a new message

{
  "message_id": "000000000000000000000001",
  "feedback": "Ask one qualifying question before suggesting a call.",
  "corrected_text": "What is your main goal for the next month?"
}

review_learning_candidates

List pending candidates[] with previews and quality scores. Approve or reject a reviewed candidate_id. Approval creates a training example.

OAuth scope · write

Arguments · action (list / approve / reject), candidate_id (string)

Supply candidate_id for approve or reject.

Return fields · total, candidates, success, candidate_id, message

Example · List pending learning candidates

{
  "action": "list"
}

get_memory

List active memories[] by type in priority order. total is the number returned. Use a selected entry’s id for deletion.

OAuth scope · read

Arguments · type (rule / correction / tone / example / knowledge / template / coaching), limit (integer)

Return fields · total, filter_type, memories, message

Example · Inspect active rules before adding or replacing one

{
  "type": "rule",
  "limit": 50
}

update_sales_rules

Creates a new rule each time. The result includes the stored rule and potential conflicts. A conflict warning still leaves the new rule in place.

OAuth scope · write

Required arguments · content (string)

Other arguments · category (objection / greeting / closing / pricing / qualifying / booking / handover / general), priority (integer)

Return fields · success, rule, conflicts, conflict_count, message

Example · Ask for the prospect’s preferred time before sharing a booking link

{
  "content": "Ask what time works for the prospect before offering a booking link.",
  "category": "booking",
  "priority": 50
}

store_knowledge

Save text in content or import a document from file_url. Choose type: knowledge for information specific to the customer. Coaching entries (type: coaching) are shared across customers. An uploaded file can still be processing after the upload succeeds.

OAuth scope · write

Arguments · content (string), file_url (string), filename (string), type (coaching / knowledge), category (objection / greeting / closing / pricing / qualifying / booking / handover / general)

Supply either content or file_url.

Return fields · success, entry, message

Example · Save the length of the onboarding call

{
  "content": "Our onboarding call lasts 30 minutes.",
  "type": "knowledge",
  "category": "general"
}

delete_memory

Deactivate the entry identified by memory_id. Winly also tries to remove it from memory search. If it was already inactive, the tool returns success: false.

OAuth scope · write

Required arguments · memory_id (string)

Return fields · success, deleted_memory, memory, message

Example · Deactivate a selected obsolete memory

{
  "memory_id": "000000000000000000000003"
}

Credits and purchases

get_credit_balance

Read workspace usage, balance and available purchases. After checkout, supply the exact checkout_session_id. purchase.status: credited confirms the grant. credits_pending means it still needs verification.

OAuth scope · read

Arguments · checkout_session_id (string)

Return fields · overview, catalog, upgrade_plan, purchase, can_continue, balance_verified, checked_at, usage_url, plan_url

Example · Check credit balance

{}

quote_credit_purchase

Quote a pack or USD amount. Show quote.credits and quote.usd for approval. Keep the returned quote_token and expires_at.

OAuth scope · read

Required arguments · mode (packs / amount)

Other arguments · pack_key (string), pack_count (integer), amount_usd (number)

For packs, supply pack_key and pack_count. For amount, supply amount_usd.

Return fields · quote, quote_token, expires_at

Example · Get the price of one credit pack

{
  "mode": "packs",
  "pack_key": "catalog_pack_key",
  "pack_count": 1
}

create_credit_checkout

Exchange the approved token for a hosted Stripe checkout_url and checkout_session_id. After a timeout, reuse the same token. Verify the specific purchase before resuming the original task with the user’s approval.

OAuth scope · write

Required arguments · quote_token (string)

Return fields · status, checkout_url, checkout_session_id, credits, usd, usd_cents, replayed

Example · Open checkout for the approved quote

{
  "quote_token": "token_from_quote_credit_purchase"
}