DeLive Transcript Analyzer

Analyze and extract insights from real-time transcription sessions captured by DeLive, a desktop app for live speech-to-text.

Prerequisites

• DeLive must be running locally (REST API at http://localhost:23456)
• For MCP integration, the DeLive MCP server must be configured (see Setup below)

Setup

Option A: MCP Server (recommended for Claude Desktop / Claude Code)

The DeLive MCP server provides direct tool access. Add to your MCP config:

{
  "mcpServers": {
    "delive": {
      "command": "node",
      "args": ["<PATH_TO_DELIVE>/mcp/delive-mcp-server.js"]
    }
  }
}

Option B: REST API (for any client)

DeLive exposes a local REST API when running:

• Base URL: http://localhost:23456/api/v1/
• WebSocket live stream: ws://localhost:23456/ws/live

Available Tools (via MCP)

Tool	Purpose
search_transcripts	Find sessions by keyword in title or transcript content
get_session	Full session with transcript, corrected transcript, AI summary, mind map, Q&A
get_session_transcript	Transcript text + corrected transcript (when available)
get_session_summary	AI summary, action items, keywords, mind map
get_recording_status	Check if DeLive is currently recording
list_topics	List topic categories for organizing sessions
list_tags	List all tags used to label sessions

Available Resources (via MCP)

Resource URI	Description
delive://sessions/recent	Most recent 10 sessions (metadata)
delive://status	Current app and recording status

Workflow Patterns

Pattern 1: Meeting Summary to Email Draft

1. Search for the relevant meeting: search_transcripts("weekly standup")
2. Get the full session: get_session("<session_id>")
3. Use the transcript and AI summary to draft a follow-up email

Pattern 2: Lecture Notes to Study Guide

1. Find the lecture: search_transcripts("machine learning lecture")
2. Get the transcript: get_session_transcript("<session_id>")
3. Extract key concepts, create flashcards, or generate a structured study guide

Pattern 3: Code Discussion to Implementation

1. Search for the discussion: search_transcripts("refactor database layer")
2. Get session details: get_session("<session_id>")
3. Extract technical decisions and action items from the summary
4. Generate implementation code based on the discussed approach

Pattern 4: Multi-Session Analysis

1. Search broadly: search_transcripts("project alpha")
2. Retrieve summaries for each matching session
3. Synthesize a cross-session report: timeline, decisions made, open items

Pattern 5: Best-Quality Transcript

1. Get the transcript: get_session_transcript("<session_id>")
2. Check if a corrected transcript is present (returned as a separate section)
3. Prefer the corrected version for downstream processing (summaries, translations, reports)

Pattern 6: Real-Time Monitoring

Connect to the live WebSocket for real-time transcript access:

import asyncio
import websockets
import json

async def monitor():
    async with websockets.connect("ws://localhost:23456/ws/live") as ws:
        async for message in ws:
            data = json.loads(message)
            if data["type"] == "transcript":
                print(data["stableText"])

asyncio.run(monitor())

REST API Reference

All endpoints return JSON. Base URL: http://localhost:23456

Method	Endpoint	Description
GET	/api/v1/health	Server health and version
GET	/api/v1/sessions	List sessions (params: search, limit, offset, topicId, status)
GET	/api/v1/sessions/:id	Full session detail
GET	/api/v1/sessions/:id/transcript	Transcript text + corrected transcript
GET	/api/v1/sessions/:id/summary	AI summary and mind map
GET	/api/v1/topics	All topics
GET	/api/v1/tags	All tags
GET	/api/v1/status	Recording state and app info

Tips

• Search is case-insensitive and matches both title and transcript content
• Sessions with status: "completed" have full transcripts; "recording" means in-progress
• The hasSummary field in session listings indicates whether AI post-processing has been run
• Use limit and offset for pagination when there are many sessions
• The live WebSocket at /ws/live broadcasts both transcript updates and session lifecycle events (session-start, session-end)
• Corrected transcript: get_session_transcript returns a correctedTranscript field when AI correction has been applied. Prefer this over the raw transcript for higher accuracy
• get_session includes a Corrected Transcript section when available — use it for summaries, reports, and analysis

Error Handling

If DeLive is not running, all API calls will fail with a connection error. Check:

1. DeLive app is open and running
2. The built-in server is active (check http://localhost:23456/api/v1/health)
3. For MCP: the MCP server process can reach DeLive on localhost