Use Visito with Codex, ChatGPT, and Claude

Connect Codex, ChatGPT, or Claude to Visito to work with your business in natural language. You can read conversations, identify common questions, and perform the operations you authorize. This guide uses a custom remote MCP connection. MCP is the connection that gives your assistant access to Visito tools. A packaged plugin can also include operating instructions, but it is not required for this setup. Visito’s public plugin-store installation is not part of this guide.

Before you start

Have an active Visito administrator account for the business you want to connect.
Use Codex with remote MCP support, ChatGPT with custom MCP access, or Claude with access to custom connectors. A managed Claude workspace may require its owner to add the connector first.
For a messaging test, use a contact and channel you are authorized to message. Reading conversations does not require sending permission.

Use this server address in any of these assistants:

https://platform-api.visitoai.com/m2m/v1/mcp

Sign in on Visito’s website when prompted. Do not paste your Visito password, API key, or OAuth callback URL into an assistant conversation.

1. Add Visito

Codex
ChatGPT
Claude

Ask Codex to help

Paste this into Codex:

Add a remote MCP server named visito with this URL:
https://platform-api.visitoai.com/m2m/v1/mcp
Use OAuth sign-in and guide me through connecting my Visito business.
Open Visito's sign-in page so I can choose my business and permissions.

Codex can help configure the server when its environment allows it. You still complete Visito sign-in and choose the permissions in the browser. If configuration is unavailable from the conversation, use the manual method below.

Add it manually

In the desktop app, open Settings → MCP servers → Add server. Name it visito, choose Streamable HTTP, and paste the server URL above. Save, select Restart when offered, then Authenticate. Complete Visito’s browser consent. These are the controls documented by OpenAI; labels can vary by app version.Alternatively, with the Codex CLI installed, run:

codex mcp add visito --url https://platform-api.visitoai.com/m2m/v1/mcp

OAuth sign-in normally opens automatically. If authentication is still required, run:

codex mcp login visito

Keep the login command running while you complete the browser flow. After authorizing in Visito, the browser returns to a temporary address on your computer. Wait for Codex to confirm successful authentication, then return to the app and open a new conversation if the tools are not yet available.

codex mcp list

This lists the configured server; the identity test below verifies that it can actually access Visito. See OpenAI’s MCP setup documentation for configuration details.

Ask ChatGPT to help

Help me connect Visito using this remote MCP server:
https://platform-api.visitoai.com/m2m/v1/mcp
Name it Visito and use OAuth. Guide me through adding a custom MCP app,
then open Visito's sign-in page so I can choose my business and permissions.

A prompt can guide setup; it does not guarantee that ChatGPT can install the connection for you. Use the manual steps if needed.

Add it manually

Open Plugins → Add → Create MCP App in ChatGPT. If custom MCP setup is unavailable, check Settings → Security and login → Developer mode and your workspace policy.
Enter Visito as the name, a short description, and the server URL above.
Choose OAuth. Let ChatGPT discover the authorization settings. In Advanced OAuth settings, keep Client Identifier Metadata Document (CIMD) when shown. You do not need to enter a client secret or the Claude client ID.
Review the requested scopes. For a small read-only test, keep only channels:read and conversations:read. If you need other operations later, request those scopes and authorize them in Visito; the consent page cannot add scopes excluded here.
Review the notice and choose Create, then Continue to Visito.
Sign in with your registered Visito admin account. Select the business and permissions as described below, then choose Connect.
Back in ChatGPT, open the installed Visito plugin and choose Try in chat. Run the quick test below.

Use Visito to identify the connected account and business, then list one
connected channel with its name and type only. Do not read customer
conversation contents, send messages, or change anything.

Verified in ChatGPT web on September 30, 2026: OAuth sign-in and a read returned the connected Visito account and an Instagram channel. The test granted only channel and conversation read access; sending and other writes were not tested in ChatGPT. Custom MCP availability and labels can vary by account or workspace. See OpenAI’s connection instructions.

Ask Claude to help

Paste this into Claude:

Help me connect Visito using this remote MCP server:
https://platform-api.visitoai.com/m2m/v1/mcp
Name it Visito. Open the connector setup if you can, or give me the
manual steps. Use OAuth sign-in; do not ask for passwords or API keys in chat.

If Visito is already configured, Claude may show a Connect or Reconnect card. A prompt alone does not guarantee installation of a new custom connector. Use the manual setup when needed; a directory result is not proof that Visito has a public store listing.

Add it manually

Open Customize → Connectors in Claude.
Open Add connector (the + menu) and choose Add custom connector.
Enter Visito as the name and the server address above as the remote MCP URL.
Choose Continue, then Sign in now.
Under OAuth client, choose Use your own OAuth client. Enter visito-claude-private-test as the client ID and leave the client secret empty.
Choose Add, complete the Visito sign-in and permission steps below, then return to Claude.
Enable Visito from the conversation’s connector controls if it is not already enabled.

The current registered public client ID has the legacy name visito-claude-private-test. It identifies the OAuth application; it is not a password, API key, or access token. Your Visito sign-in, chosen business, and granted permissions determine access. Use this predefined-client option: Claude’s published-identity and automatic-registration options are not currently registered for this integration.

See the Claude setup screens

Claude custom connector form with Visito and its public MCP URL — Add the name and remote MCP URL.

Claude OAuth configuration with the predefined public Visito client ID — Choose Sign in now and Use your own OAuth client. Leave the secret empty.

Claude Free currently allows one custom connector. If the add option is unavailable, check your plan’s limit and workspace policy before removing an existing connector. See Claude’s custom connector instructions.

The connection opens Visito’s normal sign-in page. Use your registered administrator account, including Google sign-in if that is how you normally access Visito. After login, you return to Connect Visito.

Empty Visito sign-in form with Google and email sign-in options — Visito's regular sign-in page. Spanish interface shown; the fields are the same in English.

Check the account and business before continuing. Use Switch account or Switch tenant if needed. Only eligible active businesses where you are an administrator are listed. Each connection belongs to the business selected here. Changing businesses elsewhere in the Visito dashboard does not change the connection’s business. Connect again to authorize a different business.

3. Select permissions

Every domain starts with No access. Choose only what your assistant needs:

Selection	What it allows
No access	No permissions from that domain.
Read	The available read operations for that domain.
Read & write	All available read and write actions in that domain, including sending, deletion, automation, or execution where listed.
Custom	Only the individual actions you selected under Customize permissions.

Read-only domains offer only No access and Read. Read all, Read & write all, and Clear all apply to the permissions available in the current request. Limited request means the assistant requested only part of a domain; expand Permission details to see exactly what is available.

Visito domain permissions with five selected actions and the Connect button — A narrow test selection: read channels and templates, with custom conversation permissions. Account details are excluded from the capture.

For the examples below, start with these individual permissions in Customize permissions:

Task	Permissions
Analyze conversation topics	View connected channels; Read conversations and messages.
Send a normal reply and check its result	The above, plus Send replies and attachments; Check message delivery.
Inspect approved WhatsApp templates	View WhatsApp templates. This does not grant template sending.

Keep the other domains at No access. Choose Connect when the account, business, and actions are correct. Codex, ChatGPT, and Claude may separately ask you to approve tool calls; their approval does not add permissions to the Visito connection.

4. Run a quick read test

First check the connection:

Use Visito to identify the connected account and business. Tell me what the
connection profile actually reports. Do not assume that a profile ID is a
business ID, and do not infer permissions from the list of discovered tools.
Do not change anything.

Then try a bounded conversation analysis:

Using Visito, review the 20 most recent conversations in the connected business.
Read the AI summaries first. Read up to five recent customer messages only when a summary is absent, stale, or needs supporting detail. Follow pagination
as needed and count unique conversations. What are the most common topics?
Give counts, the number of conversations successfully reviewed, and any gaps.
Avoid names, phone numbers, and identifying quotes. Do not send messages or
change any conversation. Describe this as a sample, not all account activity.

Check that the assistant reads summaries and any necessary messages instead of guessing from conversation titles. If you request a larger period, ask it to preserve the same filters while following nextCursor and check hasMore. A first page is not the entire result set, and new activity can affect a live list.

5. Send one test reply

Use your own test contact. Replace both placeholders below with a verified name and full international phone number; do not run the prompt with placeholders still present.

Find my test contact [CONTACT NAME], phone [INTERNATIONAL PHONE NUMBER],
in Visito. Verify the exact conversation and connected channel. If there are
multiple matches, ask me to choose. Check whether a normal reply is allowed.

If it is allowed, send exactly one message:
"Hello, this is a test of the Visito connection. ✅"

Do not create a new conversation, substitute a template, or send a campaign.
Keep the current AI pause state unchanged if the reply operation supports it.
Use one idempotency key for this logical send, including any safe retry.
Check the message receipt and distinguish queued, sent, delivered, and read.
If sending is not possible, explain the reason without trying another route.

A reply is a real customer message. Visito replies can temporarily pause its AI depending on the operation’s settings. Inspect the proposed recipient, channel, text, and pause behavior before approving a send.

For WhatsApp, a normal reply depends on the conversation’s messaging eligibility. Outside the permitted reply window, an approved template may be needed. Ask the assistant to inspect the channel’s available templates, language, approval, and required variables first. Template approval alone does not prove the channel is ready to send. Approve the exact template and recipient before sending, and grant template-send permission only if needed. A queued or accepted result is not delivery confirmation. Ask for the receipt status; if the provider has not returned a delivery update, the correct result is delivery pending. Repeating the request with a new idempotency key can create a duplicate message. For a campaign, start by asking for a draft audience and template plan. This connection does not expose a dedicated campaign object; do not assume that a request to “send a campaign” creates a managed campaign in Visito.

6. Manage or disconnect the connection

Open the intended business in the Visito dashboard.
Open Developers → API keys & apps. In Spanish: Desarrolladores → Claves API y apps.
Set the access-type filter to Connected applications / Aplicaciones conectadas.
Open View permissions / Ver permisos on your connection to inspect its domains, creator, and activity.
To revoke access, choose Disconnect application / Desconectar aplicación and confirm the intended connection.

Grouped permissions in Visito with a Disconnect application button, shown in Spanish — The permission drawer includes the disconnect action. This example shows a broader existing grant, not the recommended test permissions.

Disconnecting one grant leaves other connections active. If the disconnect action is unavailable, ask the connection owner or an eligible administrator to review it. To change granted permissions, disconnect and reconnect with the desired selection. Remove the plugin from ChatGPT, the connector from Claude, or the MCP configuration from Codex if you also want to remove it from that assistant. For Codex:

codex mcp logout visito
codex mcp remove visito

Revoking Visito access prevents new authenticated operations; it does not erase information already copied into an assistant conversation.

Troubleshooting

What you see	What to do
An active administrator account is required	Switch to your registered Visito administrator account and check the selected business.
Connection access unavailable / support-access session	Use Switch account and sign in with your own registered administrator account. Support-access sessions cannot authorize apps.
ChatGPT does not show Create MCP App	Check Developer mode and your account or workspace policy.
Insufficient permission or scope	Inspect the connection in Visito, then reconnect with the specific missing action. Discovering a tool does not authorize its use.
Codex returns to `127.0.0.1` and the connection is refused	Start a fresh `codex mcp login visito` and keep it running until completion. An old consent tab may point to an expired local callback listener.
Chrome shows `ERR_BLOCKED_BY_CLIENT`	Have the browser owner review the blocking extension or managed-browser policy. Restart the sign-in flow after the block is resolved.
Authentication complete, but the browser page looks plain	This final local callback page belongs to the assistant. Return to Codex and verify the connection with a read test.
Claude cannot add another custom connector	Check the plan’s connector limit or the workspace owner’s policy.
Unrecognized OAuth client in Claude	Choose Use your own OAuth client and enter the exact public client ID above. Leave the secret empty; contact Visito support if it is still rejected.
Wrong business in the result	Stop before writing, disconnect the incorrect grant, and reconnect with the intended business.

For developers, see the API documentation. Visito enforces business isolation and the connection’s granted scopes on each operation.

Actions and execution inspection

Use Actions MCP names to manage HTTP, Sheets and collection definitions. get_action_execution accepts includeInspection: true with explicit action_executions:read and action_inspection:read grants; select Execution details during consent. Saved responses may contain customer data. includeParameters and its permission remain separate. Inspection does not execute actions. Direct tests remain HTTP-only, with per-request confirmation for writes; use Playground for Sheets/collection flows. Deprecated Tools remain available with no final retirement date. See the Actions guide and migration.

AI conversation summaries

When available, list_conversations and get_conversation return the same optional aiSummary as GET /m2m/v1/conversations and GET /m2m/v1/conversations/{conversationId}. It describes the main topics and current state in one short paragraph, in the customer’s language. Summaries cover conversations independently of CRM or lead qualification.

{
  "aiSummary": {
    "text": "The guest asked about a three-night stay and airport pickup. The assistant shared room and transfer options. The guest has not selected a room or provided arrival details; no booking confirmation appears in the conversation.",
    "generatedAt": "2026-10-13T12:01:00Z",
    "throughMessageAt": "2026-10-13T12:00:00Z",
    "isStale": false
  }
}

generatedAt is when the summary was produced. throughMessageAt is the latest external message included. isStale is true while newer activity or incomplete history awaits processing. The field is omitted until a summary is available. Summaries refresh asynchronously after a short quiet period; an active conversation may temporarily have an older summary. Read summaries first for topic review and initial context. Retrieve messages when a summary is missing, stale, or you need supporting detail. A summary does not replace authoritative needsReply, booking, payment, or delivery fields. The assistant can open a conversation’s customer-facing history with list_conversation_messages. Ask it to review text, reply relationships, reactions, contacts, locations and interactive selections, as well as the available attachment metadata. Conversation lists and AI summaries are useful starting points; an absent/stale summary still requires messages for current detail. For an image, document, audio or video attachment, use get_conversation_attachment with the conversationId, stored messageId and media[].attachmentIndex from history. This uses your existing conversation-read permission. It returns metadata, processing state and existing audio/video context by default. pending, failed, timed_out and unavailable are not empty messages or successful processing. Only request includeContent: true when the actual image or file is needed. Images are returned as native MCP content; other supported files are embedded resources. Whether your assistant can display/read that content depends on its MCP host. The built-in Visito Agent stages authorized images and files in its private, network-disabled analysis workspace. It opens images with the native workspace image viewer before describing them.

{
  "conversationId": "conversation-id",
  "messageId": "000000000000000000000001",
  "attachmentIndex": 0,
  "includeContent": true
}

Inspection is limited to 2 MiB per file and a 10-second download deadline. Large/unsupported files still return metadata and an explicit status. The Agent’s document reader supports PDF, text/CSV/JSON, DOCX and XLSX with limits of 20 PDF pages, 16,000 characters, or two spreadsheet sheets/100 rows/20 columns, plus 5 CPU seconds and 512 MiB. It reports truncation and cannot extract text from encrypted/scanned documents in every case. Spreadsheet formulas are not executed. Stored audio transcripts and video context are reused first. This tool does not create new transcription, video analysis or OCR, run guest files, send messages or change conversation state. Native file/sticker message types remain readable; they do not become newly supported outbound-send formats. Refresh your connection’s cached tools when the new reader is deployed. See the message/attachment API contract for statuses and examples. Image inspection also limits dimensions to 4,096 pixels per side and 16 megapixels. Animated GIF/WebP and malformed/oversized image headers return unsupported; they are not silently interpreted as a complete animation. The metadata read remains available.

Availability and compatibility

This October 13, 2026 source update is additive and has been tested locally through authenticated API/MCP transports and with synthetic images/files in the managed Agent runtime. It does not itself deploy or enable the feature in production. Deploy the compatible API/MCP reader first, then the Agent runtime, and refresh cached tool definitions. Existing optional media outcomes/video context are read without a migration or historical repair. AI summary generation is separately controlled by its global worker switch; missing summaries remain valid responses. The read remains under existing conversations:read authority. Processing or inspection failure does not authorize sending a message, replaying ingestion, retrying a business write, or changing a customer record. External MCP hosts must support native image/file results to interpret the actual bytes. The managed Assets/Knowledge library is a separate surface.

Explain a saved AI answer

Ask the agent to inspect why a particular conversation answer was produced. It can use get_conversation_answer_trace for saved planner calls, RAG source receipts and response inputs/outputs. External connections need Conversation debugging → Read (conversations:debug:read) separately from normal conversation access; existing grants are unchanged. The built-in Visito Agent inherits active admin business access. Metadata is default; content is bounded and sanitized on explicit request. Missing/expired/partial/oversized/truncated evidence must be disclosed. Reads never replay tools or send messages. The additive reader is locally implemented; production availability requires its compatible release. See the contract and limits.

Read longer handoff histories

list_conversation_handoffs returns up to thirty total handoff/review cases per page. Follow nextCursor until hasMore=false. Each case has only its latest five events; use list_conversation_handoff_events with the case’s eventsNextCursor to continue older events (default fifty/max one hundred). Preserve cursor limits and disclose missing/oversized payloads. Current ownership comes from conversation metadata. Both tools use existing conversations:read; neither resolves a case nor sends a notification. This source update is locally verified; release requires indexes and refreshed MCP schemas. Contract.

Choose the conversation evidence tool

MCP descriptions explain when to use a tool, where its IDs come from, and which tool provides the next piece of evidence. The October 13 source update adds this routing guidance to conversation tools and carries API parameter descriptions into MCP inputs. It requires API deployment and refreshed tool discovery; descriptions do not grant permissions.

Question	Tool path	Identity and coverage
Which guest conversation is this?	`list_conversations` with `contactQuery` and `channel` → `get_conversation`	Check `participant.displayName` and optional `participant.phone`. `list_assignable_operators` identifies staff.
What are guests asking about?	`list_conversations` summaries → `list_conversation_messages` for missing/stale summaries or detail	Paginate or disclose the sample. This is not an account-wide topic aggregation or message-text search.
What was said or attached?	`list_conversation_messages` → `get_conversation_attachment`	Stored `messageId` plus `media[].attachmentIndex`; read metadata/context before opting into file content.
Why did the AI answer this way?	`list_conversation_messages` → `get_conversation_answer_trace`	Select the AI assistant’s stored `messageId`; separate debugging permission. Follow `pagination.nextOffset`; disclose capture gaps.
What action ran, and what did it return?	`list_action_executions` with `conversationId` → `get_action_execution`	Use `entry.receiptId` as `executionId`. Parameters and inspection need independent permissions; history defaults to seven days. Do not rerun actions to debug.
Why was it handed to a person?	`get_conversation` → `list_conversation_handoffs` → `list_conversation_handoff_events`	Current ownership is separate from history. Case pages and event pages have separate cursors; events are not message IDs.
Was an accepted message delivered?	`get_message_request_status` → `get_message_status`	Start with the accepted `requestEventId`, then a returned stored `messageId`. `externalMessageId` is a provider identity, not either input.
What is currently configured or booked?	`list_actions` → `get_action`; `list_conversation_reservations`	Definitions are current configuration; reservation records are mirrored context. Neither replaces a saved action result or live availability query.

Message, handoff and action reads remain separate evidence sources. There is no universal public link from every action to an exact answer; do not infer one from timestamps alone.

Knowledge and guest AI configuration through MCP

The local October 13, 2026 continuation adds existing knowledge and guest-AI configuration tools. Deploy the compatible services and refresh tool definitions before using them. Select Knowledge read/write and the separately consented AI personality read/write permissions; existing grants do not gain access automatically. Saved answer debugging still has its own permission.

Task	Tool path	What to verify
Discover knowledge	list_knowledge_items / list_knowledge_sources → get_knowledge_source	Follow nextCursor; include inactive sources for diagnosis. Title/URL query is metadata search.
Inspect indexed text	search_knowledge → get_knowledge_chunk	Current semantic search can call an embedding provider; scope with propertyIds/includeGlobal. fallback_all is bounded unranked fallback.
Inspect an uploaded original	list_knowledge_sources → get_knowledge_original_file	Private expiring URL; asset-only uploads can have no indexed text.
Create/edit text	get_knowledge_source → ingest_knowledge_text	Reuse the text sourceId and send the whole replacement; read active/quota state afterward.
Manage website knowledge	list_knowledge_websites → get_knowledge_website → create/recrawl/selection/property tools	Discover page/path IDs; read get_knowledge_crawl and actual selection/generation/quota, not only acceptance.
Manage shareable assets	list_knowledge_sources(assetsOnly=true) + list_knowledge_asset_folders → sharing/folder tools	Validation, scope, sharing and folder activation are separate from KB activation. Folder deletion removes members.
Change tone/name/instructions	get_ai_settings → update_ai_settings	Keep fields the user did not ask to change, copy updatedAt to expectedUpdatedAt; explicitly preserve or clear customInstructions.
Change photo limit	get_ai_asset_sharing_settings → update_ai_asset_sharing_settings	One–fifteen photos, default six; fresh version required.
Change escalation rules	get_ai_handoff_policy → update_ai_handoff_policy	Fresh expectedPolicyVersion, complete rule array, retain intended IDs; working hours/knowledge-gap gates still apply.

Guest personalization affects customer replies/Playground; CRM follow-up drafts use personalization through their own resolver. It does not configure Support Copilot or the separate managed Business Agent. The dashboard compiledSystemPrompt is a preview; runtimePersonalization shows stored inputs and legacy verbose. Core rules, tools, knowledge, channel/property/conversation state and server fallback can also affect a reply. For “Why did my AI answer this?”, begin with the saved message and get_conversation_answer_trace, then compare current knowledge/settings only as current diagnosis. Current text, activation, generation or timestamps cannot prove answer-time configuration. Disclose missing/expired/truncated capture and absent source/settings version linkage; do not replay tools, call providers, enable capture, send a message or change settings merely to explain a saved answer. Read the Knowledge API and AI settings contract for limits, permissions and mutation semantics. After a 409 version conflict, reload and reconcile. After an ambiguous ingestion/crawl/deletion outcome, discover state before retrying. Ingestion can save inactive knowledge because of quota; accepted/queued does not mean ready or relevant.

Channel connection links

To add a Meta channel, ask the connected agent to read connection options and create a temporary channel connection link. Open it yourself, sign in as an administrator of the intended business, and complete provider consent/resource selection. Instagram, Messenger and WhatsApp have different browser steps. Ask the agent to check the saved attempt and resulting channel readiness before declaring success. Links expire in 15 minutes and require the separately granted connection permission. Webchat requires snippet installation; TikTok remains a gated native setup. See channel tools and recovery. This addition requires the compatible release; links are disabled until enabled by the installation.

Ask about how Visito works

With knowledge:read permission, the assistant can use search_visito_docs to find published product/API guides in English or Spanish, then read_visito_doc to read a matching page. Ask, for example: “How do handoff required fields work? Explain using the Visito documentation and include source links.” These tools retrieve documentation as needed; they do not load a fixed manual or read your business’s Knowledge Base. Documentation explains product behavior; separate authorized business reads establish your actual settings. Results include source links and disclose truncation. Newly published changes become available after documentation indexing and a cache of up to 60 seconds. These reference reads have no additional Lana action charge; ordinary turn and business action charges still apply. The two read tools require compatible platform and Lana releases. The separate public documentation MCP is https://docs.visitoai.com/mcp; Visito exposes only search and reading, without documentation feedback or arbitrary commands.

Find chats by availability or booking actions

For an action-based export, ask the agent to use list_conversations with actionId/actionStatus before reading messages. This selects recent chats by projected history. For actions executed during a specific period, use business-wide list_action_executions with explicit dates and omit conversationId; paginate and deduplicate matches before reading their contact metadata. Checkout completion is not confirmed payment/booking, and missing history limits coverage. Examples and selection semantics. The new conversation filters require API release and refreshed tools.

Manage tickets

Explicitly grant Tickets access and any required escalation creation/management permissions. Follow discover → inspect evidence → select a human → start → add notes → resolve → read back. Customer replies use separate sending tools.

Business overview

Use get_business_overview to orient the agent before business-context work. It summarizes the connected business, saved channels and integration/action capabilities, knowledge and assets, guest AI/handoff configuration, sales CRM, products/services, the last 30 days of activity, and current attention counts. No additional OAuth permission is required. Each section uses its existing read permission; unauthorized, unavailable and incomplete data are explicit and never represented as zero. The connected business comes from authentication, not tool arguments. Full prompts, customer messages and credentials are excluded. Results carry timestamps and a cache expiry (60 seconds). Pass refresh: true after relevant configuration changes. Saved readiness does not establish live provider health or permission to execute; follow the returned inspection tools and existing action approvals. This helper adds no Lana action credit. The underlying authenticated API is GET /business/overview?refresh=true.

Manage Commerce work

Use list_tickets(includeCommerce=true, source=commerce), inspect get_ticket, select a human owner, Start work and add notes. With explicitly granted sales read and the dedicated write scope, call update_commerce_sale_fulfillment or reconcile_commerce_sale using a fresh sourceVersion and Idempotency-Key, then read back. Reconciliation is recorded evidence, not provider confirmation. Generic Resolve/Cancel cannot bypass Commerce completion. Refunds, payouts and booking retries remain human actions.

Getting Started

General

Lana

CRM

Agent

Best Practices

Connections

Tools

Workspace

Use Visito with Codex, ChatGPT, and Claude

Before you start

1. Add Visito

Ask Codex to help

Add it manually

Ask ChatGPT to help

Add it manually

Ask Claude to help

Add it manually

3. Select permissions

4. Run a quick read test

5. Send one test reply

6. Manage or disconnect the connection

Troubleshooting

Actions and execution inspection

AI conversation summaries

Read message attachments

Availability and compatibility

Explain a saved AI answer

Read longer handoff histories

Choose the conversation evidence tool

Knowledge and guest AI configuration through MCP

Channel connection links

Ask about how Visito works

Find chats by availability or booking actions

Manage tickets

Business overview

Manage Commerce work

​Before you start

​1. Add Visito

​Ask Codex to help

​Add it manually

​Ask ChatGPT to help

​Add it manually

​Ask Claude to help

​Add it manually

​2. Sign in and choose your business

​3. Select permissions

​4. Run a quick read test

​5. Send one test reply

​6. Manage or disconnect the connection

​Troubleshooting

​Actions and execution inspection

​AI conversation summaries

​Read message attachments

​Availability and compatibility

​Explain a saved AI answer

​Read longer handoff histories

​Choose the conversation evidence tool

​Knowledge and guest AI configuration through MCP

​Channel connection links

​Ask about how Visito works

​Find chats by availability or booking actions

​Manage tickets

​Business overview

​Manage Commerce work

Before you start

1. Add Visito

Ask Codex to help

Add it manually

Ask ChatGPT to help

Add it manually

Ask Claude to help

Add it manually

2. Sign in and choose your business

3. Select permissions

4. Run a quick read test

5. Send one test reply

6. Manage or disconnect the connection

Troubleshooting

Actions and execution inspection

AI conversation summaries

Read message attachments

Availability and compatibility

Explain a saved AI answer

Read longer handoff histories

Choose the conversation evidence tool

Knowledge and guest AI configuration through MCP

Channel connection links

Ask about how Visito works

Find chats by availability or booking actions

Manage tickets

Business overview

Manage Commerce work