discord-rag

name	discord-rag
description	Ingest channel messages into LanceDB and run semantic search (discord rag_ingest / rag_query).

Discord MCP — RAG over messages

Description: Turn Discord message history into a searchable knowledge base. Covers message indexing, semantic search across channels, user activity pattern extraction, topic clustering, and cross-channel knowledge retrieval.

Trigger Phrases

• "Search past Discord messages about [topic]"
• "What was decided about [feature] in [channel]?"
• "Find all messages from [user] about [subject]"
• "Summarize the discussion in [channel] from last week"
• "What support questions were asked about [product]?"
• "Index channel [name] for semantic search"

Tools

• discord(operation='list_channels', guild_id='...') — List available channels in a guild. Get channel_id for targeting.
• discord(operation='get_messages', channel_id='...', limit=50) — Raw message fetch without RAG.
• discord(operation='rag_ingest', channel_id='...', limit=100, guild_name='...', channel_name='...', table_name='discord_messages') — Ingest messages into LanceDB vector index. Embeds message content, author, and timestamp.
• discord(operation='rag_query', query_text='...', top_k=10, table_name='discord_messages') — Semantic search over ingested messages. Returns content + author + timestamp + channel context.

Workflow

1. Channel discovery: Call discord(operation='list_channels', guild_id='...') to find the target channel. Note the channel_id.
2. Ingest: Call discord(operation='rag_ingest', channel_id='...', limit=50) to index recent messages. For full history, repeat with paginated limit calls.
3. Query: Use discord(operation='rag_query', query_text='...', top_k=10) for semantic search. Prefer narrow, specific queries for best results.
4. Synthesis: When returning results, cite channel, author, and timestamp. Summarize repeated themes across results.

Notes

• Ingest pulls text via the same path as get_messages; large limits increase API load (rate limits apply).
• Embeddings use local sentence-transformers + LanceDB paths under the server process. No external API calls.
• Ingestion is incremental — re-ingesting the same channel with a higher limit adds new messages without duplicating.
• Prefer narrow queries; cite channel and message context when summarizing hits.
• table_name parameter allows multiple named indexes (e.g., per guild or per topic).

Examples

• "Find discussion about the new authentication flow." → rag_query(query_text="new authentication flow", top_k=10)
• "Index the #support channel for RAG." → rag_ingest(channel_id="123456", limit=200, guild_name="My Guild", channel_name="support")
• "What did Alice say about deployment last month?" → rag_query(query_text="Alice deployment", top_k=5)