Grounded Knowledge Engine MCPサーバー

Grounded Knowledge Engine

Local-first, provider-neutral project memory for AI coding agents.

Grounded Knowledge Engine (GKE) gives Claude Code, Codex, Gemini CLI, and other MCP clients durable context across sessions. It retrieves evidence from local files, answers with file-and-line citations, retains useful learning back into Markdown, and resumes structured project state without turning a database or hosted service into the source of truth.

Your Markdown remains canonical. Retrieval indexes are derived and disposable, and every workspace stays on your machine.

https://github.com/user-attachments/assets/d00c977b-9538-4835-94d5-26350136eb34

What GKE does

• Grounded answers with file-and-line citations and strict abstention.
• Durable capture back into Markdown; risky changes enter a review queue.
• Project resume as a cited briefing: focus, decisions, blockers, next steps.
• Document ingestion for PDF, DOCX, XLSX, PPTX, HTML, and more.
• One local MCP server for Claude Code, Codex, Gemini CLI, and Copilot.

It is not a hosted chat memory, team wiki, or document-management platform.

Quick start

GKE requires Node 22.5 or newer; Node 24 is recommended.

Install the current release:

npm install --global https://github.com/dimosgit/grounded-knowledge-engine/releases/download/v0.2.2/grounded-knowledge-engine-0.2.2.tgz

Create the verified demo workspace and configure supported clients:

gke demo
cd gke-demo
gke setup

Restart Claude Code or Codex from gke-demo, then ask:

Use GKE to resume the router-rollout project.

Continue with the five-minute golden path to retrieve cited evidence, retain one learning, and retrieve it again from a fresh session.

The core loop

Index or ingest local evidence, answer with citations, capture what was learned back into Markdown, and resume the project in a later session.

MCP surface

The default core profile exposes four tools: kb.search, kb.get_record, kb.answer_and_capture, and kb.resume_project. When writes are enabled, retention is automatic and update-first: when the agent establishes a finding the knowledge base lacked, it is appended to the record that owns it, without asking first. The demo and default workspaces enable writes; named workspaces start with writes disabled. The full profile adds advanced retrieval, decisions, and explicit write tools. See the MCP server reference.

Your own workspace and documents

gke setup --workspace client-alpha --workspace-root "/path/to/client-alpha"
npm run ingest -- ./inbox --dry-run

Each named workspace runs as its own process, with writes disabled by default. See workspace configuration and the ingestion guide.

Operator Cockpit

An optional local view for project attention, capture and decision review, and a context graph. Try the public demo or read the Cockpit guide.

Learn more

Golden path · Boundaries · Data model · Copilot · Agent workflows · Contributing

License

MIT. The demo knowledge base includes paraphrased notes from the MIT-licensed MCP documentation; see demo sources.