AgencyQ Client Knowledge Engine

A deployed, multi-tenant knowledge service that gives AI agents deep, queryable understanding of client content, brand, and data — exposed via authenticated MCP.

Why

AI agents working on client projects lack structured access to client knowledge. They don't know the brand voice, can't look up real stats, and invent content instead of using what the client provided. This tool fixes that.

How It Works

AI Agent ──MCP──→ Knowledge Engine ──→ PostgreSQL + pgvector
                                        (Supabase)

Three layers of client intelligence:

1. Session Brief — client context loaded at session start via MCP resource/prompt
2. Semantic Search — natural language queries over all client documents
3. Structured Records — typed, extractable data (stats, colors, partners)

MCP Tools

| Tool | Purpose | |------|---------| | search_knowledge | Semantic search over client documents | | get_client_brief | Concise client overview for passive context | | get_brand_voice | Detailed voice guidelines and example copy | | get_structured_data | Typed records (stats, colors, partners) | | ingest_content | Add curated content to the knowledge base | | ingest_file | Queue raw files for extraction (v0.5+) | | list_knowledge | Inventory of ingested content |

Stack

• MCP Server: Local Node.js (v0.1), Cloudflare Worker (v0.5+)
• Database: PostgreSQL + pgvector on Supabase
• Embeddings: OpenAI text-embedding-3-small
• Auth: API key per developer, scoped to client(s)

Status

Pre-development. PRD complete at docs/prd.md.

Phased Rollout

• v0.1 — Local server + Supabase, NGS as client zero (2-3 days)
• v0.5 — Deployed to Supabase + Cloudflare, multi-client (1-2 weeks)
• v1.0 — Full ingestion pipeline, CLI, monitoring (2-4 weeks)