Agentibus MCP connector

Agentibus OÜ · Documentation

Last updated: 28 September 2026

Agentibus runs AI sales agents for businesses on Vendus, its sales-agent platform. The agents reply to leads on WhatsApp, Instagram Direct, Messenger, web chat and email, qualify them and book sales calls into the business's calendar. Businesses manage their agents from the dashboard at app.vendus.ai. Vendus is made by Agentibus OÜ.

The Agentibus MCP connector lets an AI assistant that supports the Model Context Protocol (MCP), such as Muse, Claude or ChatGPT, work inside your Vendus workspace on your behalf: read leads, conversations and results, change how your agents behave, and approve or send replies. Using the connector requires an Agentibus account; if you don't have one yet, signing in with Google puts you on the early-access list.

New to Agentibus?

Get your own AI sales agent for Instagram and WhatsApp leads: it replies to every lead, qualifies them and books your sales calls.

Build your agent in 60 seconds, free

Want it inside Claude? Connect Agentibus and sign in with Google to join the early-access list.

1. Connection details

Item	Value
Endpoint	`https://vendus-mcp-directory.fly.dev/mcp`
Transport	MCP Streamable HTTP, stateless
Authentication	OAuth 2.1 authorization code flow with PKCE (S256). Clients can identify themselves with a Client ID Metadata Document (the HTTPS URL of their metadata as client_id) or register with Dynamic Client Registration (RFC 7591). No API keys are accepted on this endpoint.
Authorization server metadata	`https://api.vendus.ai/.well-known/oauth-authorization-server`
Protected resource metadata	`https://vendus-mcp-directory.fly.dev/.well-known/oauth-protected-resource/mcp`
Scopes	`platform:read`, `platform:write`
Rate limit	120 requests per minute per connection
Account	An Agentibus customer account (the Vendus dashboard login) with access to the workspace. New to Agentibus? Your first 10 booked sales calls are free: start at agentibus.ai

2. Authorization and permissions

2.1 How you connect

Your AI assistant reads the protected resource metadata, finds the Vendus authorization server and registers itself as an OAuth client.
You are sent to a consent page on app.vendus.ai. You sign in with your Vendus dashboard login and see which application is asking for access, for which workspace and with which scopes. You approve or deny.
If you approve, the assistant receives an access token bound to that one workspace. It acts as you, only on that workspace and only with the scopes you granted.
Access tokens expire after 1 hour. The assistant renews them with a refresh token, which expires after 30 days or earlier if you revoke the connection.

2.2 Scopes

Scope	What it allows
`platform:read`	Reading data of the workspace: agents, leads, conversations, replies awaiting approval, knowledge base, triggers, reminders and metrics.
`platform:write`	Changing the workspace and acting in it: agent and lead modes, knowledge base, approving and sending replies, discarding drafts, cancelling reminders.

Every request is checked against the granted scopes: reading needs platform:read, anything else needs platform:write. The test_agent tool sends nothing to anyone, but it runs the agent, so it needs platform:write.

2.3 Controls

Workspace binding: a token works only on the workspace chosen at consent. Requests for any other workspace are refused.
Audit log: every call, including refused ones, is written to an audit log with time, workspace, user, application and the API route called.
Rate limit: 120 requests per minute per connection. Requests above the limit are refused.
Revocation: you can revoke a connection at any time in the Vendus dashboard, under Settings > API Keys, in the list of MCP connections. Its tokens stop working immediately.
Workspace switch: the connector can be turned off for a whole workspace. When it is off, every connector token for that workspace is refused. The workspace owner can ask for this at [email protected].

3. Tools

The connector exposes 19 tools. Each one is marked with the kind of effect it has:

Read Reads data. Changes nothing.
Write Changes settings or content in your workspace. Contacts no one, and can be undone.
Sends / irreversible Sends a message to a real person, or removes something that cannot be restored.

3.1 Read (12)

Tool	What it does	Type
`whoami`	Returns the workspace, the signed-in user, the granted scopes and the workspace's agents. Assistants are told to call it first.	Read
`list_agents`	Lists the AI agents of the workspace with id, name, channel and status.	Read
`get_agent_mode`	Shows the current mode of an agent: autopilot, hybrid or off.	Read
`list_leads`	Lists leads, most recent activity first. Supports search by name or phone and pagination.	Read
`get_conversation`	Returns the full conversation history with one lead.	Read
`list_pending_replies`	Lists the reply drafts waiting for human approval (hybrid mode).	Read
`test_agent`	Chats with an agent in a sandbox to see how it would answer. Nothing is sent to any lead.	Read
`list_knowledge_documents`	Lists the knowledge base documents (files and web pages) and their indexing status.	Read
`list_triggers`	Lists the events that make an agent start a conversation.	Read
`get_metrics`	Returns funnel, daily volume and lead counts for a period.	Read
`get_business_metrics`	Returns replies, bookings and time saved for one agent.	Read
`list_reminders_queue`	Lists the scheduled follow-up reminders.	Read

3.2 Write, reversible (4)

Tool	What it does	Type
`set_agent_mode`	Sets an agent to autopilot (replies on its own), hybrid (every reply waits for approval) or off (does not reply).	Write
`set_lead_ai_mode`	Overrides the mode for a single lead, or removes the override so the lead follows the agent again.	Write
`add_knowledge_url`	Adds a web page to the knowledge base of the workspace or of one agent.	Write
`add_knowledge_text`	Adds a text document (for example an FAQ or a price list) to the knowledge base.	Write

3.3 Sends or irreversible (3)

Tool	What it does	Type
`approve_pending_reply`	Approves a waiting draft and sends it to the lead, optionally with edited text. A sent message cannot be recalled.	Sends / irreversible
`reject_pending_reply`	Discards a waiting draft. Nothing is sent to the lead, and the discarded draft is not kept for approval.	Sends / irreversible
`cancel_reminders`	Cancels every scheduled follow-up reminder for one lead.	Sends / irreversible

These three tools ask for confirmation on the server. Called without confirm=true, they change nothing and return a preview of the exact action: who receives the message and its text, the draft that would be discarded, or the reminders that would be cancelled. The assistant shows you the preview and calls the tool again with confirm=true only after you approve that specific action.

Agents can be referred to by name or id. When no agent is given, tools use the default agent of the workspace. The full list of tools, with their input schemas, is also returned by the server when your assistant connects.

4. Example prompts

“How many sales calls did my agent book this week?”
“Show me the replies waiting for my approval”
“Pause my Instagram agent until Monday”
“What did leads ask about most in the last 7 days?”
“Add https://example.com/pricing to my agent's knowledge base”

5. How to connect

Client	How
Muse	Once Agentibus is listed in the Muse connector directory, add it from there. You are sent to the Agentibus consent page to sign in and approve access.
Claude.ai	Settings > Connectors > Add custom connector, then enter `https://vendus-mcp-directory.fly.dev/mcp`.
ChatGPT	Settings > Apps & Connectors, then add a remote MCP server with the URL `https://vendus-mcp-directory.fly.dev/mcp`.
Claude Code	Run the command below, then run `/mcp` inside Claude Code to sign in.
Other MCP clients	Any client that supports remote MCP servers with OAuth can connect with the same URL.

claude mcp add --transport http vendus https://vendus-mcp-directory.fly.dev/mcp

6. Safety notes

Writes are real. Approving a reply delivers a message to a real lead, and it cannot be recalled. Changing an agent's or a lead's mode changes how the agent talks to real leads from the next message.
Nothing irreversible happens without your approval. Sending a reply, discarding a draft and cancelling reminders first return a preview and run only when called again with confirm=true after you approve. The server also instructs assistants to confirm with you before changing an agent's or a lead's mode. Read what the assistant proposes before you say yes.
Hybrid mode keeps a human in the loop. In hybrid mode every reply the agent writes waits for approval, in the dashboard or through approve_pending_reply, before it reaches the lead.
Check the consent page. Approve access only for the workspace and scopes you intend to use, and revoke connections you no longer use.

7. Data handling

The connector server holds no credentials of its own and stores nothing. It forwards your OAuth access token to the Vendus API over TLS. The Vendus API checks the token, the workspace, the scope and the rate limit on every request.
The data you reach through the tools is your business's own lead data, which Vendus already processes as a processor for your business under the Data Processing Agreement.
Vendus receives only what your assistant sends in a tool call: the tool name and its arguments. It does not receive the rest of your conversation with the assistant.
Tool results are returned to the AI assistant you use, which handles them under its provider's own terms and privacy policy.

Details are in the connector Privacy Policy. Use of the connector is governed by the connector Terms of Use.

8. Support and contact

Questions, access problems and requests: [email protected].

Security reports: [email protected].