Knowledge-Server

Persistent graph memory for AI agents.

Knowledge-Server is an MCP (Model Context Protocol) server that gives AI agents persistent, queryable memory using Neo4j. Your agents remember conversations, learn patterns, track decisions, and build knowledge over time.

Why Knowledge-Server?

AI agents have a memory problem. Context windows aren't memory—they're short-term buffers that vanish between sessions. Knowledge-Server solves this by providing:

Graph-native memory — Relationships are first-class, not an afterthought
Semantic search — Find memories by meaning, not just keywords
Session continuity — Pick up exactly where you left off
Execution planning — Track multi-step workflows with dependencies
Issue tracking — Log problems and link them to solutions

Features

Memory Tools

store_fact — Store knowledge with tags and confidence scores
store_conversation_memory — Preserve conversation exchanges
query_memory — Text search across all memories
semantic_query_memory — Find conceptually similar content
get_context_map — Navigate related knowledge

Planning Tools

create_plan — Define execution plans with task dependencies
get_next_tasks — Get tasks ready to execute
track_task_progress — Real-time progress updates

Session Tools

create_session — Start a work session with objectives
update_session_progress — Track accomplishments and blockers
complete_session — Hand off context to future sessions

Issue Tools

create_issue — Log problems with severity
link_issue_to_plan — Connect issues to remediation plans

Quick Start

1. Prerequisites

Python 3.10+
Neo4j database (local or cloud)
Claude Code or another MCP client

2. Install

git clone https://github.com/tonton2006/knowledge-server.git
cd knowledge-server
pip install -r requirements.txt

3. Configure

cp .env.example .env
# Edit .env with your Neo4j credentials

4. Run

python main.py

5. Connect from Claude Code

Add to your Claude Code MCP settings:

{
  "mcpServers": {
    "knowledge-server": {
      "command": "python",
      "args": ["/path/to/knowledge-server/main.py"],
      "env": {
        "NEO4J_URI": "bolt://localhost:7687",
        "NEO4J_USER": "neo4j",
        "NEO4J_PASSWORD": "your-password"
      }
    }
  }
}

Example Usage

Store a fact

store_fact(
    fact_text="User prefers dark mode interfaces",
    category="preference",
    tags=["user", "ui", "settings"]
)

Query by meaning

semantic_query_memory("What are the user's display preferences?")
# Finds the dark mode fact even though words don't match

Start a work session

create_session(
    date_str="2026-03-02",
    focus="Implement authentication system",
    planned_objectives=["Design schema", "Write endpoints", "Add tests"]
)

Architecture

┌─────────────────┐     ┌─────────────────┐
│   MCP Client    │────▶│ Knowledge-Server │
│  (Claude Code)  │◀────│    (FastMCP)     │
└─────────────────┘     └────────┬────────┘
                                 │
                        ┌────────▼────────┐
                        │     Neo4j       │
                        │ (Graph Database)│
                        └─────────────────┘

Configuration

Variable	Description	Default
`NEO4J_URI`	Neo4j connection URI	`bolt://localhost:7687`
`NEO4J_USER`	Neo4j username	`neo4j`
`NEO4J_PASSWORD`	Neo4j password	(required)
`GCP_PROJECT_ID`	For Vertex AI embeddings	(optional)
`OPENAI_API_KEY`	For OpenAI embeddings	(optional)
`LOG_LEVEL`	Logging verbosity	`INFO`

Pricing

Free: 50,000 facts, basic features
Pro ($99/mo): 500,000 facts, semantic search, priority support
Enterprise: Custom limits, SLA, dedicated support

License

MIT License — see LICENSE for details.

Contributing

Contributions welcome! Please read our contributing guidelines before submitting PRs.

Knowledge-Server