What is MCP, in plain terms?

The Model Context Protocol is an open standard that lets an AI assistant use an outside service. Connecting WhisperChat's MCP server means your assistant can build and manage your chatbots directly, instead of you clicking through the dashboard.

Do I need an API key?

No. The MCP server uses OAuth, so you sign in to WhisperChat in your browser and approve the connection. No key is ever copied into your AI client, and you can revoke access from your client's connector settings whenever you like.

Can a connected assistant see my other chatbots or customers?

It can only ever see chatbots on your own account, because every request is made as you and ownership is checked on each call. It cannot reach anyone else's chatbots or data.

My client says the server needs authentication. What do I do?

That is the normal first step — the server refuses anonymous requests by design. Look for a Connect, Authenticate or Needs login action next to the server in your client and follow it. In Claude Code, run /mcp and choose Authenticate.

What is the difference between search and query_chatbot?

search looks across the knowledge sources you have added and returns matching records to read. query_chatbot actually asks a chatbot the question and returns the answer it would give a real visitor, which is what you want when testing a bot.

My client isn't listed. Can I still connect it?

Almost certainly. Any client that supports remote MCP servers over HTTP with OAuth can connect — paste the server URL wherever it asks for a custom or remote MCP server. The platforms above are just the ones with steps written out. Watch the transport field, as its name differs per client: type http in VS Code, streamableHttp in Cline, streamable-http in Roo Code, httpUrl in Gemini CLI, serverUrl in Devin Desktop, and no field at all in Cursor, Zed or LM Studio.

MCP Server | WhisperChat — Connect to Claude, ChatGPT, Cursor & More

Paid plans only

Your chatbots, inside your assistant

WhisperChat runs a Model Context Protocol server. Connect it once and your AI assistant can create chatbots, import your website, train them and test their answers — by just asking. No API keys to manage: you sign in with WhisperChat and approve the connection.

Paid plan required

The MCP server is available on paid subscriptions only. Free accounts can't authorize a connection or create chatbots through it — approving the connection prompts you to upgrade if your plan isn't active. Your plan's normal message and character limits still apply.

View plans

See it in action

Watch an assistant create a chatbot, crawl a site and train it — end to end, just by asking.

Add to Claude Other platforms

Setup by platform

Every client below speaks remote MCP over HTTP. Pick yours for exact steps.

Claude

Claude web and desktop, via custom connectors.

Add to Claude

Before you start

Available on Free, Pro, Max, Team and Enterprise. Free accounts can add one custom connector.
On Team and Enterprise, an owner or admin has to add the connector for the organisation first.

1
Use the Add to Claude button above — it opens Claude's add-connector dialog with the WhisperChat URL already filled in. To do it by hand instead, follow the steps below.
2
Open Claude's connector settings.
claude.ai/customize/connectors
3
Click the + button, then choose Add custom connector.
4
Name it WhisperChat and paste the server URL. Leave the Advanced OAuth fields empty — the server registers your client automatically.
Server URL
```
https://www.whisperchat.ai/mcp
```
5
Click Add, then Connect. Claude sends you to WhisperChat to sign in and approve the permissions, then returns you with the tools ready.

Custom connectors are in beta. Claude reviews a connector before it can be used, so a prefilled link never skips that step.

Claude connects from Anthropic's servers rather than from your computer, so this works with the hosted WhisperChat server and not with a local dev URL.

Claude Code

The terminal agent — one command.

Before you start

Claude Code installed and signed in.

1
Add the server. --scope user makes it available in every project; leave it off to keep it in the current project's .mcp.json only.
Terminal
```
claude mcp add --transport http --scope user whisperchat https://www.whisperchat.ai/mcp
```
2
Sign in. Run /mcp inside Claude Code, pick WhisperChat and choose Authenticate — or do it straight from the terminal.
Terminal
```
claude mcp login whisperchat
```
3
Confirm it connected.
Terminal
```
claude mcp list
```

If you write .mcp.json by hand, a remote server needs "type": "http" alongside its url. Without the type it is read as a local command and silently skipped.

ChatGPT

Custom connectors, behind Developer Mode.

Before you start

A Plus, Pro, Business, Enterprise or Education plan. Custom connectors are not on the Free tier.
ChatGPT on the web. The setup flow isn't available in the mobile apps.
On Business and Enterprise, an admin may need to allow Developer Mode first.

1
Open Settings → Security and login and turn on Developer mode.
chatgpt.com/settings
2
Go to Settings → Plugins and choose the + button to create one.
chatgpt.com/plugins
3
Give it a name and description, paste the server URL, and set authentication to OAuth.
Server URL
```
https://www.whisperchat.ai/mcp
```
4
Click Create, then connect — you'll be sent to WhisperChat to sign in and approve access. The new connector appears under Drafts.
5
In a chat, open the + menu and choose More to enable WhisperChat for that conversation.

OpenAI publishes no deep link for custom connectors, so the URL has to be pasted.

Deep Research needs a server to expose tools named exactly search and fetch. WhisperChat provides both — search finds your knowledge sources, fetch reads one in full. Ordinary Developer Mode connectors don't need them.

Cursor

The AI editor — one-click install.

Add to Cursor

Before you start

Cursor installed on this machine, since the button opens it via a cursor:// link.

1
Use the Add to Cursor button above, then click Install in the dialog Cursor opens.
2
Prefer to edit the config? Open Settings → MCP → Add new MCP server, or edit ~/.cursor/mcp.json directly and add:
~/.cursor/mcp.json
```
{
  "mcpServers": {
    "whisperchat": {
      "url": "https://www.whisperchat.ai/mcp"
    }
  }
}
```
3
Open Settings → MCP and click Needs login (or Connect) next to WhisperChat to sign in and approve access.

Cursor identifies a remote server by the presence of url — there is no type field here, unlike VS Code.

VS Code

Copilot agent mode, with one-click install.

Add to VS Code

Before you start

VS Code with GitHub Copilot enabled, and Copilot's agent mode.

1
Use the Add to VS Code button above and confirm the install prompt.
2
Or run MCP: Add Server from the Command Palette, choose HTTP, and paste the server URL.
Server URL
```
https://www.whisperchat.ai/mcp
```
3
To edit it by hand, put this in .vscode/mcp.json for one project, or your user mcp.json for all of them. Note the top-level key is servers, not mcpServers.
.vscode/mcp.json
```
{
  "servers": {
    "whisperchat": {
      "type": "http",
      "url": "https://www.whisperchat.ai/mcp"
    }
  }
}
```
4
Open the Copilot chat view in agent mode. VS Code prompts you to sign in to WhisperChat the first time a tool is used.

VS Code handles the OAuth flow for you — there's nothing to configure beyond the URL.

Grok

Custom connectors on grok.com.

Before you start

A Grok account. For a team, Team Read-Write access in the xAI console.

1
Open Grok's connectors page.
grok.com/connectors
2
Choose New Connector, then Custom.
3
Name it WhisperChat and paste the server URL.
Server URL
```
https://www.whisperchat.ai/mcp
```
4
Save, then complete the sign-in and approval flow in Grok.

For a team account, add it from console.x.ai under Connectors → Add Connector → Other instead.

xAI's API supports remote MCP too, but only with a static token — it can't complete the OAuth sign-in that WhisperChat requires. Use the Grok app for this connection.

Gemini

The Gemini app — custom apps for Spark.

Before you start

A personal Google Account. Custom apps don't work with a work or school account.
You must be 18 or over and in the US, and the feature is only available in English.
Access to Gemini Spark, and Keep Activity switched on — Gemini can't connect to custom apps without it.

1
In the Gemini app, open Settings & help → Connected Apps.
2
Under Custom apps for Spark, choose Add a custom app.
3
Paste the server URL and choose Next. Leave the Advanced features credentials alone — WhisperChat registers your client automatically.
Server URL
```
https://www.whisperchat.ai/mcp
```
4
Sign in to WhisperChat and approve the permissions.

The eligibility rules here are strict and set by Google, not by WhisperChat. If you use a work Google Account, the Gemini CLI is the way in instead.

Gemini CLI

Google's terminal agent.

Before you start

Gemini CLI installed.

1
Add the server. The --transport flag matters: it defaults to stdio, which is for local programs.
Terminal
```
gemini mcp add --transport http whisperchat https://www.whisperchat.ai/mcp
```
2
Sign in from inside the CLI. Your browser opens to approve access.
In Gemini CLI
```
/mcp auth whisperchat
```
3
If you edit settings.json by hand, use httpUrl. The plain url key means SSE, which this server doesn't serve.
settings.json
```
{
  "mcpServers": {
    "whisperchat": {
      "httpUrl": "https://www.whisperchat.ai/mcp"
    }
  }
}
```

Mistral Vibe

Custom MCP connectors in Vibe (formerly Le Chat).

Before you start

Administrator access. On Free, Pro and Student plans the account owner is the administrator, so this is usually just you.

1
In Vibe, switch to Work mode and open the Connectors page.
2
Click + Add Connector and switch to the Custom MCP Connector tab.
3
Paste the server URL and choose OAuth 2.1 with dynamic client registration — WhisperChat supports it, so there are no credentials to enter.
Server URL
```
https://www.whisperchat.ai/mcp
```
4
Save, then sign in to WhisperChat and approve access.

Le Chat was renamed Vibe, and MCP connectors live under Work mode.

LM Studio

Local models, with a one-click install link.

Add to LM Studio

Before you start

LM Studio installed on this machine.

1
Use the Add to LM Studio button above and confirm the install dialog.
2
Or edit ~/.lmstudio/mcp.json by hand. LM Studio follows Cursor's notation, so a remote server is just a url.
~/.lmstudio/mcp.json
```
{
  "mcpServers": {
    "whisperchat": {
      "url": "https://www.whisperchat.ai/mcp"
    }
  }
}
```
3
LM Studio opens your browser to sign in to WhisperChat when the server is first used.

On Windows the file lives at %USERPROFILE%/.lmstudio/mcp.json.

Zed

Native MCP with automatic sign-in.

Before you start

A recent Zed build with the agent panel.

1
Open Settings → AI → MCP Servers, click Add Server, then choose Add Remote Server.
2
Or add it to settings.json directly. Zed calls these context_servers rather than mcpServers.
settings.json
```
{
  "context_servers": {
    "whisperchat": {
      "url": "https://www.whisperchat.ai/mcp"
    }
  }
}
```
3
Leave out any Authorization header. Zed treats a remote server without one as needing OAuth and prompts you to sign in.

Cline

The VS Code agent extension.

Before you start

The Cline extension in VS Code, or the Cline CLI.

1
In the extension, open the MCP Servers panel and use the Configure tab.

For the CLI, edit ~/.cline/mcp.json. Cline spells the remote transport streamableHttp, in camelCase.

~/.cline/mcp.json

{
  "mcpServers": {
    "whisperchat": {
      "type": "streamableHttp",
      "url": "https://www.whisperchat.ai/mcp"
    }
  }
}

3
Save. Cline connects and prompts you to authenticate in the browser.

Roo Code

Nearly Cline's format — with one key difference.

Before you start

The Roo Code extension installed.

1
Edit mcp_settings.json for every project, or .roo/mcp.json for just this one. Note the transport is streamable-http with a hyphen — Cline spells the same thing streamableHttp.
.roo/mcp.json
```
{
  "mcpServers": {
    "whisperchat": {
      "type": "streamable-http",
      "url": "https://www.whisperchat.ai/mcp"
    }
  }
}
```
2
Save, then authenticate when Roo Code prompts you.

Devin Desktop

Formerly Windsurf.

Before you start

Devin Desktop (the app previously called Windsurf).

1
Edit the MCP config and add WhisperChat under serverUrl. The config path still uses the old Windsurf name.
~/.codeium/windsurf/mcp_config.json
```
{
  "mcpServers": {
    "whisperchat": {
      "serverUrl": "https://www.whisperchat.ai/mcp"
    }
  }
}
```
2
Reload the MCP servers in Cascade, then sign in when prompted.

A plain url key works here too — serverUrl is just what the docs show.

Copilot Studio

Add WhisperChat as a tool for your agents.

Before you start

A Copilot Studio agent with generative orchestration turned on — MCP tools are ignored without it.

1
Open your agent and go to Tools → Add a tool → New tool.
2
Select Model Context Protocol to open the onboarding wizard.
3
Give it a name and description and paste the server URL.
Server URL
```
https://www.whisperchat.ai/mcp
```
4
For authentication choose OAuth 2.0 → Dynamic discovery. WhisperChat supports dynamic client registration, so there's nothing to paste.
5
Click Create, then add the tool to your agent and sign in when prompted.

MCP tools ride on Power Platform connectors, so your organisation's DLP policies apply to them.

Claude Desktop (legacy)

Only if you can't use the Connectors UI.

Before you start

Node.js installed. This route runs a local bridge process with npx.

1
Try the Connectors UI first. Claude Desktop adds remote servers natively through Settings → Connectors, with no bridge and no config file — see the Claude tab above. Only read on if that isn't an option for you.

Open Settings → Developer → Edit Config to reveal claude_desktop_config.json, then add:

claude_desktop_config.json

{
  "mcpServers": {
    "whisperchat": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://www.whisperchat.ai/mcp"]
    }
  }
}

3
Save and restart Claude Desktop fully. A browser window opens to sign in and approve.

The mcp-remote bridge is needed because this config file only speaks to local programs — it has no field for a remote URL. The Connectors UI has no such limit.

What your assistant can do

Once connected, these tools are available. Your assistant picks the right one — you just describe what you want.

list_chatbotsread

List every chatbot on your account, with its ID.

get_chatbotread

Read one chatbot's full settings.

create_chatbotwrite

Create a new chatbot.

update_chatbotwrite

Rename a bot, switch its model, or restyle the whole widget.

delete_chatbotdelete

Permanently delete a chatbot.

add_knowledge_sourceknowledge

Add one page, text snippet or Q&A pair.

crawl_websiteknowledge

Crawl an entire site into a bot's sources.

train_chatbotwrite

Train on the current sources. Runs in the background.

get_chatbot_statusread

Check training progress for each source.

query_chatbotread

Ask a chatbot a question and get its real answer.

get_embed_coderead

Get the snippet to embed the widget on your site.

searchread

Search across your chatbots' knowledge sources.

fetchread

Read one knowledge source in full, by ID.

Try asking: “Create a chatbot called Support Bot, crawl example.com into it, train it, then ask it what our refund policy is.” Your assistant will chain create_chatbot → crawl_website → train_chatbot → query_chatbot on its own.

Sign-in and permissions

WhisperChat's MCP server uses OAuth 2.1 with PKCE. There is no API key to copy, paste or leak — your client is sent to WhisperChat to sign in, and you approve exactly what it may do.

How connecting works

1Your client discovers the server and registers itself automatically.
2It opens WhisperChat in your browser, where you sign in as normal.
3You review the permissions requested and approve the connection. A paid plan is required at this step — free accounts are asked to upgrade before a connection is granted.
4The client receives an access token valid for one hour, refreshed silently for 30 days.

Permissions you grant

chatbot:read

List your chatbots, read their settings and ask them questions.

chatbot:write

Create chatbots, change their settings and start training.

chatbot:delete

Delete chatbots. Granted separately so a mistake can't remove your work.

knowledge:write

Add web pages, text and Q&A pairs to a chatbot's knowledge.

A connected client can only ever see chatbots on your own account. Revoke access at any time from your client's connector settings.

REST API docs Setup guide

WhisperChat MCP Server

Your chatbots, inside your assistant

Setup by platform

Claude

Claude Code

ChatGPT

Cursor

VS Code

Grok

Gemini

Gemini CLI

Mistral Vibe

LM Studio

Zed

Cline

Roo Code

Devin Desktop

Copilot Studio

Claude Desktop (legacy)

What your assistant can do

Sign-in and permissions

How connecting works

Permissions you grant

MCP Questions

WhisperChat MCP Server

Your chatbots, inside your assistant

Setup by platform

Claude

Claude Code

ChatGPT

Cursor

VS Code

Grok

Gemini

Gemini CLI

Mistral Vibe

LM Studio

Zed

Cline

Roo Code

Devin Desktop

Copilot Studio

Claude Desktop (legacy)

What your assistant can do

Sign-in and permissions

How connecting works

Permissions you grant

MCP Questions

What is MCP, in plain terms?

Do I need an API key?

Which WhisperChat plan do I need?

Can a connected assistant see my other chatbots or customers?

My client says the server needs authentication. What do I do?

What is the difference between search and query_chatbot?

My client isn't listed. Can I still connect it?