WeSpeak MCP Connection

WeSpeak MCP is WeSpeak's official connector for Claude (claude.ai): it adds to your Claude the same tools as the assistant you already use in the panel, so you can run your account by talking to your own AI — search conversations, move leads, check metrics, edit your assistant's knowledge or pause it — without opening the panel.

MCP (Model Context Protocol) is the standard protocol that AIs like Claude use to connect to external tools. Since the WeSpeak tools live alongside whatever else you have connected, you can ask Claude for things the panel can't do: for example, a report that combines your WeSpeak metrics with your spreadsheets.

In this article

• What you need
• How to connect it
• What each role can do
• Use cases: what you can ask it for
• What it can modify (and what it can't)
• Security and privacy
• If you manage several properties
• Troubleshooting
• Current limits
• Frequently asked questions

What you need

1. The feature enabled on your account — your WeSpeak advisor handles this. Once enabled, your connection details appear on the Settings screen of the panel.
2. A Claude account (claude.ai) that allows adding connectors.
3. Your usual WeSpeak user. There are no new keys or API keys: access is authorized by signing in, and every action respects your user's role.

How to connect it

1. In the WeSpeak panel, go to Settings: you'll see the Connect WeSpeak to Claude section, with your details ready to copy — the suggested connector name, your account's URL and the Client ID — plus the step-by-step video.
2. In claude.ai, go to Settings → Connectors → Add custom connector.
3. Give it the suggested name (you can change it) and paste the server URL exactly as copied. It's unique to your account — it ends in ?username= followed by your user — and it's what identifies your connection: don't trim it or use another account's.
4. In the connector's advanced settings, paste the Client ID (a public identifier, not a password). Don't skip it: without it the connection fails.
5. Add the connector: it shows up in your connectors list. Click Connect on the WeSpeak one — the WeSpeak sign-in opens: log in with your usual user and authorize access. This step can take up to 30 seconds; wait for Claude to redirect you back.
6. Done — in a new chat, Claude has your WeSpeak tools. Try "find yesterday's conversations that were left unanswered".

ℹ️ Each person on the team connects their own user from their own Claude. Don't share users: that way every action is recorded under the name of whoever did it.

What each role can do

The rules are the same as in the panel: each user only sees the tools their role allows, and higher roles add to the lower ones.

ℹ️ In the connector's tool list you'll see technical names in English (searchConversations, knowledgeAssistant, testAssistant…). No need to memorize them: Claude picks the tool based on what you ask for.

Use cases: what you can ask it for

Claude doesn't just run one tool at a time: it chains several within a single request (search → read → summarize → act) and writes the result itself in whatever format you ask for — a list, a table, a report. These are real examples of what you can ask it, grouped by task; in brackets, the minimum role you need.

Day-to-day conversation handling (Operator · Supervisor to act)

It searches by text, status (open/closed), channel (WhatsApp, Instagram, Facebook), assigned agent and date range, and reads a conversation's full thread — including transcriptions of audio messages that were already generated in the panel.

• "Show me today's open conversations that still have no agent assigned."
• "Find this week's conversations where guests ask about late check-out or early check-in."
• "Summarize the conversation with +1 305 … and tell me what's still waiting for a reply."
• "Which Instagram conversations from the weekend are still open?"
• With the Supervisor role it also acts: "Assign yesterday's open conversations about groups to Lucía", "Close last week's conversations that already have a confirmed reservation", "Snooze this conversation until Monday at 9".

💡 For bulk actions (assigning or closing several), ask for the list first and only then for the action: that way you review exactly what it's going to touch.

Leads and sales funnel (Supervisor)

It lists leads filtered by text, status, days without activity and funnel stage, and moves them between stages (won, lost with a reason, or to a specific stage).

• "List the leads that have had no activity for more than 5 days."
• "How many leads are in 'Quoted' today, and who are they?"
• "Mark García's lead as lost, reason 'chose another hotel'."
• "Build a table with the open leads, their stage and last activity so I can hand it to the sales team."
• "Give me the stalled leads, read their conversations and tell me which ones to follow up first and with what message." — Claude reads each thread, prioritizes and drafts the text; you send it from the panel.

Reports and statistics (Supervisor)

The metrics Claude can query today are two: volume (messages and conversations over a period, in days — the last week, the last 30 days…) and funnel (how many leads are in each stage and how many were won or lost in the period, with each stage's name). Everything else it builds by counting and cross-referencing the results of conversation and lead searches. Some reports you can ask for as-is:

• Weekly activity report: "Put together this week's summary: conversations and messages over the last 7 days compared with the previous week, plus the 5 most repeated questions."
• Funnel report: "How is funnel conversion looking over the last 30 days? How many leads came in, which stage is each in, and how many did we close?"
• Split by channel and by agent: "Of this week's conversations, how many came through each channel and how many does each agent have assigned?"
• Team backlog: "How many open conversations does each agent have, and which ones have gone the longest without an update?"
• Recurring topics and knowledge gaps: "Read the conversations from the last 3 days and tell me which questions repeat and in which ones the assistant handed over to an advisor." — ideal for deciding what to add to the knowledge.
• Automatic follow-ups: "How many of this week's conversations were sent the automatic follow-up message?"
• Stalled leads by stage: "How many leads have had no activity for more than 7 days in each funnel stage."
• Format and export: "Turn that report into a table I can paste into a spreadsheet", "Write it up as a short report to email to management". If you have other tools connected in your Claude (spreadsheets, documents, messaging), you can ask it to save or share it there directly.

ℹ️ Today's metrics do not include response times or the split between assistant and human replies. Reports built by listing conversations are paginated: for large volumes, narrow the period (a week, a month) so they stay accurate.

Reservations (Operator · requires WeSpeak Payments)

• "Show me the latest reservations and their status."
• "Look up reservation ABC123 and tell me what was paid and what's outstanding."
• "Did the guest of reservation ABC123 write anything after booking?" — cross-references the reservation with its conversation.

Assistant knowledge (Administrator)

It reads the knowledge by field or by semantic search, edits it with validation, backup and verification, and can undo. It's the way to keep what the virtual assistant answers up to date.

• Audit: "What does the knowledge say today about pets, parking and check-in time?", "Search the knowledge for everything related to cancellations and tell me whether there are contradictions."
• Update facts: "Change breakfast hours to 7:30–10:30", "Update the booking link", "Add the instructions for paying by bank transfer."
• Behavior rules: "Don't promise to notify the guest when a payment is verified", "Always ask for the dates and number of guests before quoting."
• When to hand over to a human: "When guests ask about events or weddings, have the assistant hand over to an advisor."
• History and undo: "What changes were made to the knowledge this week, and by whom?", "Undo the last change."
• Several complexes: say which one ("…only in the beach complex") or ask it to apply the change to all.

ℹ️ Promotions and discounts, room descriptions and photos, and answering hours are not edited from here: Claude will point you to where each one is configured (see the next section).

Assistant quality control (Administrator)

With test messages, Claude can pose as a guest, even across several turns, and compare what the assistant answers with what it should answer.

• "Ask the assistant whether pets are allowed and, in the same chat, the price for 2 people from October 12 to 15. Tell me if it answered correctly."
• "Try these 10 frequently asked questions and flag the ones where the answer was incomplete or wrong" — and then "fix the knowledge for the ones that failed and test again".
• Before the season: "Check that the assistant answers correctly about hours, policies, payment methods and directions, and fix whatever is outdated."
• Diagnosing an answer: "In this conversation the assistant said the pool is heated and it isn't: why did it say that?" — you get a verdict (missing information in the knowledge, a real error, not an error…) and, depending on the case, the fix or a report to our team.

Configuration and status (Administrator)

• "Is the assistant active? What hours does it answer?"
• "Pause the assistant" / "Resume it" — it affects every guest of the account, so Claude confirms with you first.
• Answering hours can be checked from here but are changed on the panel's Settings screen.

Help about WeSpeak (all roles)

• "How do I set up follow-up messages?", "What does Multi-account include?" — Claude searches these guides and answers with a link to the article.

Tips for asking well

• Be specific about period, channel and status: "this week's open ones on Instagram" gets better results than "the conversations".
• List first, act second for any bulk change.
• Name the property if you have several accounts connected.
• Ask for the format you need (table, list, short report) and in which language.
• If something can't be done from Claude, it will tell you and point you to the panel or your advisor.

What it can modify (and what it can't)

Most tools are read-only. From Claude you can only:

• Pause or resume the assistant. It's the only editable part of its configuration: business hours and time zone can be viewed, but they're changed in the panel.
• Edit the assistant's knowledge. Every change is validated before being applied, keeps a backup and can be undone.
• Change a conversation's status (assign, close, reopen or snooze it). It never edits the content of the messages.
• Move a lead between stages. Funnel stages can be viewed, but their structure isn't edited from here.
• Report a problem to our team when an assistant answer went wrong.

Test messages are simulations: they aren't stored and don't reach anyone. And the connector doesn't send messages to your guests: it's for running your account, not a service channel — your WhatsApp assistant keeps working exactly as always.

⚠️ Important: Claude acts on what you ask, with no extra confirmation step from WeSpeak in between. Review what it proposes before telling it to apply — especially when editing knowledge. As a safety net, those changes keep a backup and can be undone.

Security and privacy

• Your usual permissions: Claude can do exactly what you can do in the panel with your user — no more, no less — and only sees your account's data.
• Everything is recorded under your name: knowledge edits and reports register who made them.
• Your conversation is private: WeSpeak doesn't see what you talk about with your Claude or what Claude answers; it only receives the actions Claude runs on your account.
• Revocable anytime: remove the connector from Claude's settings and your user's access is cut off. If you want to disable the feature for the whole account, ask your advisor.

If you manage several properties

Each connection is tied to one WeSpeak account. To manage several properties, add one connection per account: each account has its own connection URL (the trailing ?username= tells them apart, and Claude treats each URL as a separate connector), which you copy from the Settings screen of that account's panel. They all live in the same Claude, each announces itself with its account's name and only sees its own data. When asking Claude for something, say which property it's about, so it uses the right connection.

Troubleshooting

Current limits

• Today it works with Claude (claude.ai). ChatGPT can't connect for now — a technical piece is missing on our side for its login to work — and Gemini doesn't allow connecting this kind of tool. Other MCP-compatible apps may need to be enabled by us first: check with your advisor, mentioning which app you use.
• It doesn't message guests and doesn't change how your assistant handles WhatsApp.
• It's a young conversational tool. If a tool answers oddly or a request doesn't come out as expected, report it to your advisor: it helps us improve it.

Frequently asked questions

Does it replace the panel assistant?
No, they coexist: the same tools remain available in the panel and, on top of that, in your Claude. Use whichever is more comfortable for each task.

Does it have an extra cost? Who pays for the AI?
The AI usage runs on your Claude subscription — WeSpeak doesn't charge you per query. Check with your advisor whether the feature is included in your plan.

Can Claude answer my guests?
No. The connector operates your account (conversations, leads, metrics, knowledge), but it doesn't write to anyone. That said: editing the knowledge from here does affect what your assistant answers — just like editing it from the panel.

Can several people on my team connect?
Yes: each one connects their own Claude with their own WeSpeak user and sees the tools of their role.

Does what I do from Claude count the same as doing it in the panel?
Yes — and it's recorded under your name. Claude acts on your account based on what you ask, so review what it proposes before applying it; knowledge changes keep a backup and can be undone.

How do I disconnect it?
Remove the WeSpeak connector from Claude's settings and your user's access is cut off. If you want to disable the feature for your whole account, ask your advisor.