Delive Transcript Analyzer

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

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

Pattern 2: Lecture Notes to Study Guide

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

Pattern 3: Code Discussion to Implementation

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

Pattern 4: Multi-Session Analysis

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

Pattern 5: Best-Quality Transcript

• Get the transcript: get_session_transcript("<session_id>")
• Check if a corrected transcript is present (returned as a separate section)
• 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:

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