MCP App Store
by KeenEthics
AssistLoop icon

AssistLoop MCP Server for ChatGPT

By AssistLoop

Overview

AssistLoop gives your business an AI support agent. It answers your customers on your website, WhatsApp, Instagram, Messenger, Telegram and email, and passes the conversation to a person on your team when one is needed. It doesn't stop at answers. Connect AssistLoop to the tools your business already runs on, and your AI agent can get things done for your customers right in the conversation: book or move an appointment, check on an order or delivery, update their details, or open a request. You decide exactly what it is allowed to do. It also collects leads, shares your Calendly booking link and adds tickets to HubSpot. With this plugin, your team can run all of that from ChatGPT: - Conversations: see what customers are asking, reply to them, hand a chat to your team or back to the AI agent, or close it. - Support tickets: see, create and update tickets, leave notes for your team, and email the customer. - Training: teach your AI agent from text, questions and answers, or pages from your website, and see when it has learned them. - Settings: rename an agent, choose its AI model, update its instructions, and decide when chats go to a person. Who it is for: owners, admins and members of an AssistLoop workspace. You sign in with your AssistLoop account and approve the connection, and you only see what you can see in AssistLoop. Members only see the agents they are assigned to. Good to know: the plugin cannot delete anything, create new agents, upload files, or connect your other tools; you do those in AssistLoop. Replies and emails reach real customers, so the plugin is set up to send only text you wrote or approved. Requires an AssistLoop account. Privacy: https://assistloop.ai/legal/privacy

How to connect

  1. ChatGPT: Open the AssistLoop listing in the ChatGPT app directory and choose Connect. ChatGPT listing AssistLoop exposes 31 tools here.

Tools

add_qa_source

ChatGPT
Add a question-and-answer training source to an agent: one answer, and the ways customers ask for it (at least one question). The agent learns it in the background: the result says queued, then check get_training_status with the source_id. Each call adds a new source. Send an idempotency_key so a retry cannot apply this twice.

add_text_source

ChatGPT
Add a text training source to an agent, for example opening hours, policies or product facts. The agent learns it in the background: the result says queued, then check get_training_status with the source_id. Each call adds a new source (it does not edit one). Send an idempotency_key so a retry cannot apply this twice.

add_ticket_comment

ChatGPT
Add an internal note to a ticket, for the team only: the customer never sees it (use reply_to_ticket_customer to write to the customer). body is plain text or markdown. Teammates who follow the ticket are notified. Returns the note as list_ticket_comments shows it. Send an idempotency_key so a retry cannot apply this twice.

create_ticket

ChatGPT
Create a support ticket on an agent. Teammates who follow tickets are notified; the customer is not contacted (use reply_to_ticket_customer for that). description is plain text or markdown. assigned_to is me or a teammate id from list_ticket_assignees. conversation_id links a conversation of the same agent. Returns the ticket as get_ticket shows it. Send an idempotency_key so a retry cannot apply this twice.

get_agent

ChatGPT
Get one AI agent: name, description, model (id, name, slug), handoff settings (human_intervention: visitors can ask for a person; only_human: the AI never answers), whether it is disabled and why, its organization_id and whether ticketing is on.

get_conversation

ChatGPT
Get one conversation: its id, status (open/closed), response_mode (ai/human), channel origin, the customer, and the last message. Use list_conversation_messages for the full thread.

get_crawl_results

ChatGPT
Get a website crawl's state and results: status (pending, processing, completed, failed) and, once completed, the pages found (url, already_in_library, selectable, upgrade_required, is_new since the last crawl of the site), skipped urls, plan limits and a folder tree of the pages. A crawl runs in the background; call again while it is pending or processing.

get_ticket

ChatGPT
Get one support ticket: subject, description, status, priority, requester, creator, assignee, dates, and the linked conversation (its id, status, customer and latest message). Use list_ticket_messages for the customer thread and list_ticket_comments for the team's internal notes.

get_training_progress

ChatGPT
List one agent's running and recent (last 24 hours, at most 20) website page fetches, newest first. Each has a task_id, phase (fetching, training, done, failed), whether training was requested, the page urls, the urls that failed and a failure code. Use list_sources or get_training_status to see the resulting sources.

get_training_status

ChatGPT
Get the training state of one source: training_status (pending: sent for training or not yet sent; trained; failed), training_in_progress (a website page retraining while its previous version stays live), awaiting_training_result (any source sent for training whose result has not come back; retrain_source refuses it until then), and how many of its items were processed or failed. Training runs in the background; call again later to see it change.

get_website_source

ChatGPT
Get one website page source: its url, title, the page text stored for training (cut after 4000 characters; content_truncated says when, content_length is the whole length), whether text is stored, whether the stored text changed since it was trained (content_stale), when it was fetched, and content_hash, which update_website_source_content needs. Only for sources of type url from list_sources.

list_agents

ChatGPT
List the AI agents you can see, across your premium organizations (agents of a free organization are not listed: no tool can act on them). Members see only the agents assigned to them. Each item has the agent id, name, description, model, handoff settings (human_intervention, only_human), whether it is disabled, its organization_id and whether ticketing is on. Filter with organization_id and status (active or disabled). Next: get_agent, list_conversations, list_tickets or list_sources with an agent_id.

list_ai_models

ChatGPT
List the AI models an agent can use, newest first. Each item has the model id, name, description, provider, slug, whether it is a premium model, whether it supports function calling, and the message credits one reply costs. Use a model id to compare with an agent's model from get_agent.

list_conversation_messages

ChatGPT
List the messages of one conversation, newest first: role (visitor, assistant, human), content, attachments and time. Pass meta.next_cursor back as cursor to read older messages; has_more is false on the last page.

list_conversations

ChatGPT
List the conversations of one agent, most recently active first. Each item has the conversation id, status (open/closed), response_mode (ai/human), channel origin, the customer, and the last message. Use get_conversation or list_conversation_messages for the full thread.

list_organizations

ChatGPT
List the organizations you belong to, newest membership first. Each item has the organization id, name, slug, your role (owner, admin or member), the plan name, and premium: only premium organizations can be used through this connector. Next: list_agents with an organization_id.

list_sources

ChatGPT
List the training sources of one agent (text, Q&A, website pages, files), oldest first. Each item has the source id, type, title, training_status (pending/trained/failed), training_in_progress, awaiting_training_result, item counts and size. Use get_training_status for one source, get_website_source for a website page's text, and get_training_progress for running website fetches.

list_ticket_assignees

ChatGPT
List the teammates who can be assigned this agent's tickets: id, name, email and role. Use an id as assigned_to in list_tickets.

list_ticket_comments

ChatGPT
List the team's internal notes on one ticket, newest first: body (HTML), author and dates. The customer never sees these; for the customer thread use list_ticket_messages.

list_ticket_messages

ChatGPT
List the customer thread of one ticket, oldest first: messages from the customer and the team (kind "message", body is HTML) and status changes (kind "activity"). The customer sees these messages; for the team's internal notes use list_ticket_comments.

list_tickets

ChatGPT
List the support tickets of one agent, newest first. Each ticket has its id, number (T-1001), subject, status, priority, requester, assignee, linked conversation id and whether the customer wrote since you last read it. Filter by status, priority, assigned_to (me, unassigned, or a user id from list_ticket_assignees) or a subject search. Use get_ticket, list_ticket_messages and list_ticket_comments for one ticket.

reply_to_conversation

ChatGPT
Send a reply to the customer of a conversation, as a human teammate. It is delivered to the customer right away (chat widget, WhatsApp, Telegram, Instagram, Messenger or Email) and switches the conversation to human mode, so the AI stops answering. Plain text only. Read the conversation first with list_conversation_messages; a sent message cannot be taken back. Returns the message as list_conversation_messages shows it. Send an idempotency_key so a retry cannot apply this twice.

reply_to_ticket_customer

ChatGPT
Email the ticket's customer, as a teammate: the message is stored on the ticket thread and emailed to the requester right away, with a link to answer. body is plain text or markdown. The ticket needs a requester email (see get_ticket). Read the thread first with list_ticket_messages; a sent email cannot be taken back. Returns the message as list_ticket_messages shows it. Send an idempotency_key so a retry cannot apply this twice.

retrain_source

ChatGPT
Train one source again, for example after update_website_source_content or when get_training_status says failed. Training runs in the background: the result says queued, then check get_training_status. A source still waiting for its training result (awaiting_training_result in get_training_status) is refused; wait for it to finish. Send an idempotency_key so a retry cannot apply this twice.

start_website_crawl

ChatGPT
Start crawling a website for one agent, to find its pages. It runs in the background: call get_crawl_results with the returned crawl_task_id until its status is completed, then pick pages for train_website_pages. If this agent is already crawling the same page, that crawl's crawl_task_id is returned (status already_running) instead of starting another. Public http(s) addresses only. Send an idempotency_key so a retry cannot apply this twice.

train_website_pages

ChatGPT
Fetch website pages and train the agent on them (the dashboard's Save and train). Pass the page urls, usually from get_crawl_results (at most 100). It runs in the background: follow it with get_training_progress (the task_id), then list_sources. Pages already saved are refused unless replace_existing is true. Public http(s) addresses only. Send an idempotency_key so a retry cannot apply this twice.

update_agent

ChatGPT
Change an agent's settings: its name, its AI model (an id from list_ai_models), human_intervention (visitors can ask for a person) and only_human (the AI never answers). Send only what changes; at least one. Name and model need an owner or admin. Answers the agent as get_agent shows it. Send at least one of: name, ai_model_id, human_intervention, only_human.

update_agent_prompt

ChatGPT
Replace an agent's system prompt: the instructions the AI follows in every conversation. The old prompt is overwritten and cannot be restored here, so read it in the dashboard first if it must be kept. Owners and admins only. Takes effect on the next message. Answers the agent as get_agent shows it. Send an idempotency_key so a retry cannot apply this twice.

update_conversation

ChatGPT
Open or close a conversation, or switch who answers it: response_mode "human" stops the AI so teammates reply, "ai" hands it back to the AI. Send status and/or response_mode. A closed conversation must be reopened (status "open") to change its mode. Answers the conversation as get_conversation shows it. Send at least one of: status, response_mode.

update_ticket

ChatGPT
Change a ticket's status (open, in_progress, resolved, closed), priority, assignee or subject. Send only what changes. assigned_to is me, unassigned, or a teammate id from list_ticket_assignees. Teammates who follow the ticket are notified; the customer is not. Returns the ticket as get_ticket shows it. Send at least one of: status, priority, assigned_to, subject. Send an idempotency_key so a retry cannot apply this twice.

update_website_source_content

ChatGPT
Replace the whole stored text of a website page source (type url from list_sources), for example to remove navigation or fix a wrong fact, and optionally its title. Read it first with get_website_source and send its content_hash as base_content_hash: the edit is refused if the page changed since. A page longer than get_website_source shows (content_truncated) cannot be edited here, since the rest of it would be lost; edit it in the dashboard. Answers the page as get_website_source shows it. This only stores the text: run retrain_source on the same source for the agent to learn it. Send an idempotency_key so a retry cannot apply this twice.

Capabilities

Writes

Example Prompts

Click any prompt to copy it.

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/assistloop"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/assistloop)