Uniphore Conversation AI MCP. Extract actions, summaries, and insights from calls.

Uniphore Conversation AI: Meeting Insights MCP Server

Your agent handles complex call intelligence. You can treat recordings like a structured database, pulling out specific data points using natural language prompts—no human has to listen to the audio first.

Finding Your Data

To start, you'll need to know what data exists. Use list_meetings to pull up every recorded conversation and call ID available in your organization. If you know roughly when or who was involved, you can narrow things down fast by using search_meetings, which filters past records based on keywords, specific topics, or a date range.

You don't even need to guess what kind of data is available; run list_topics and you'll get an index of every keyword or topic tracked across all recordings. When you know the exact meeting ID, you can get basic details about that single call using get_meeting.

Raw Content Extraction

When you need to see everything that was said, run get_transcript. This outputs the full, speaker-tagged text transcript for an entire recording, so every word spoken is accounted for. For a quick read of what actually happened, use get_meeting_summary to get a concise, AI-written overview of all the main topics discussed during that call.

Deep Dive Analytics and Action Items

This server pulls out structure from unstructured audio. It doesn't just give you text; it gives you actionable data points. To pull out next steps—the tasks and deliverables—run get_action_items. This tool analyzes the meeting content specifically to isolate assignees, specific tasks, and due dates. You'll get a clean list of everything that needs doing.

If you need to know how well people talked or what the general vibe was, use get_meeting_analytics. This generates detailed conversation metrics for you, including sentiment analysis (telling you if the call was positive, negative, or neutral), talk ratios between participants, and a breakdown of primary topics discussed. It’s a full performance audit of the conversation.

How it Works with Your Agent

Your agent manages this flow. You point your AI client at the server, give it the necessary access credentials for Uniphore, and then ask direct questions about your recorded calls. The agent executes the tool—say, get_action_items on a specific meeting ID—and brings back structured JSON data. This means you don't have to parse through walls of text; you just get the list: 'Task A assigned to John by Friday,' or 'Sentiment was negative during the budget discussion.' You’ll find it makes reviewing massive amounts of recorded calls fast and precise.