Elfsight AI Chatbot

What Elfsight AI Chatbot MCP server does

The Elfsight AI Chatbot MCP server gives MCP-compatible agents access to data from AI Chatbot widgets in an Elfsight account. It is hosted by Elfsight, so this repository contains documentation and a registry manifest rather than an application to build or run.

Access is scoped to the authenticated Elfsight account. An agent can work with every AI Chatbot widget in that account, but the server does not expose other Elfsight products or data. The service is suited to operational review, lead inspection, conversation analysis, and knowledge-base maintenance.

How it works

The service uses a remote Streamable HTTP MCP endpoint at https://widget-data.service.elfsight.com/mcp/ai_chat. An MCP client connects to that URL and starts an OAuth sign-in flow. The user authenticates in a browser and approves access; no API key needs to be copied into the client configuration.

The Elfsight AI Chatbot MCP server returns tool results to the connected agent. Those results may then be sent to the configured AI provider as part of the agent conversation, so returned data should be treated as potentially containing personal information.

Setup and configuration

Add the server URL to an MCP client that supports remote MCP connections with OAuth, dynamic client registration, and PKCE. Claude Code is the only client identified as tested in the supplied documentation. Its command is:

claude mcp add --transport http elfsight-ai-chatbot https://widget-data.service.elfsight.com/mcp/ai_chat

After adding it, run /mcp in Claude Code and choose authentication for elfsight-ai-chatbot. The documentation also provides URL-based configuration examples for Cursor, VS Code, and Windsurf, but states that those examples have not been verified. No environment variables or local service process are required by the documented setup.

Tools and capabilities

The available operations cover the main data types in an Elfsight AI Chatbot account:

• list_assistants lists the account’s AI Chatbot widgets.
• get_analytics retrieves performance information for a selected period.
• search_conversations finds conversations, with pagination support.
• get_conversation retrieves messages from a conversation.
• list_leads lists leads, including filters for recent or unreviewed leads.
• search_knowledge searches the chatbot knowledge base.
• list_knowledge_gaps finds visitor questions the assistant could not answer.
• answer_knowledge_gap adds a knowledge-base answer and closes a gap.

The last operation is the only documented write action. The server also includes ready-made prompts named weekly_report, knowledge_gaps_review, and lead_followup for recurring workflows.

Limitations and notes

Searches for conversations, leads, knowledge, and knowledge gaps return 20 items by default and 100 at most. Conversation and lead searches use cursors for additional pages; the other searches should be narrowed when results are truncated. Conversation retrieval returns up to 100 recent messages by default and 500 at most.

Supported reporting windows are rolling 7d, 30d, and 90d periods. Arbitrary date ranges are not available. Lead listing can also omit a period to search the full history. Rate limits apply per account and client; a 429 response includes Retry-After for retry timing.

Lead names, email addresses, and phone numbers are masked by default. get_conversation and list_leads accept contacts: "full" for an individual call. Message text, page URLs, and attachment links are not masked, and attachments are returned as links rather than file contents. Test and internal conversations are not automatically excluded. The Elfsight AI Chatbot MCP server should therefore be used with appropriate care when personal data may be present.