Knowledge Forest MCP server

Knowledge Forest MCP — Personal Knowledge & Learning Memory

A local-first personal knowledge management (PKM) and learning-memory MCP server for any AI tutor.

Turn goals into prerequisite-aware knowledge trees, preserve learning history, and require real evidence before claiming mastery.

中文 · MCP Registry · Product principles · Architecture · Contributing

Is this the MCP server you are looking for?

Choose Knowledge Forest when you want an AI tutor or learning assistant to remember more than chat history:

• Personal knowledge management: keep learner-owned goals, concepts, notes, sources, and evidence in portable local JSON.
• Learning memory across chats: let Claude, Codex, or another MCP host resume from the same durable learning state.
• A reusable knowledge graph: share one canonical concept and its learning history across multiple goals.
• Adaptive learning: expose prerequisites, blocked concepts, desired depth, and the next actionable learning node.
• Evidence-based mastery: require novel, unassisted, closed-book performance before a concept becomes verified.

This is not a vector database, document RAG server, or general-purpose transcript memory. It models what a person is trying to learn, how concepts depend on one another, and what evidence actually demonstrates mastery.

Why it exists

AI tutors are excellent at explaining a topic and terrible at owning long-lived learning state. A chat can sound productive while forgetting prerequisites, duplicating concepts across projects, or treating a polished answer as mastery.

Knowledge Forest gives the model a durable learning layer:

• Goal trees connect an observable outcome to the minimum knowledge needed to reach it.
• Canonical nodes let one concept serve several goals without copying its learning history.
• Evidence-gated mastery distinguishes reading and assisted practice from novel, closed-book performance.
• Local-first storage keeps the learner's goals, notes, and evidence in a portable JSON file.
• Model-neutral MCP works with any compatible host. The host's model does the reasoning and pays its own token cost; this server does not call an LLM.

Quick start

Requirements: Node.js 22 or newer.

Run from GitHub in any stdio MCP host

npx --yes github:znecho9/knowledge-forest-mcp doctor
npx --yes github:znecho9/knowledge-forest-mcp

Install as an MCP Bundle

Clients that support MCPB can install the self-contained bundle from the latest GitHub release. The bundle includes the server and its runtime dependencies; Node.js 22 or newer is still required.

Develop locally

git clone https://github.com/znecho9/knowledge-forest-mcp.git
cd knowledge-forest-mcp
npm ci
npm run check
npm run dev

The default data file is ~/.knowledge-forest/knowledge-forest.json. Override it with KNOWLEDGEFORESTFILE or --data-file.

Connect an MCP host

Use the exact data path you want the host to access.

Codex

Add this to ~/.codex/config.toml:

[mcp_servers.knowledge-forest]
command = "npx"
args = ["--yes", "github:znecho9/knowledge-forest-mcp"]
env = { KNOWLEDGE_FOREST_FILE = "/absolute/path/to/knowledge-forest.json" }

Claude Desktop and JSON-configured hosts

Shortened. The full README is on GitHub.