mcpField
N
neurakeep

neurakeep

Local
github.com/dominiclachanceMemory & Knowledge13Updated 4d ago

About neurakeep

neurakeep is a local-first memory system for AI agents, giving MCP clients shared memories with source citations. It preserves original materials and keeps structured memory proposals reviewable before they become durable.

Key features

  1. 01Import Markdown, text, JSONL, and PDFs
  2. 02Preserve sources and create cited proposals
  3. 03Review proposals before applying durable memory
  4. 04Search facts, events, failures, and decisions
  5. 05Explore memories in a graph
  6. 06Use a local CLI and web interface

Use cases

  1. 01Collect local documents for agent retrieval
  2. 02Review source-cited memory proposals
  3. 03Search past events and decisions
  4. 04Explore related memories through the graph

MCP support

Its local MCP server connects over stdio or HTTP and exposes tools for memory search, reading, source explanations, saving, and proposing or applying updates.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config
{
"mcpServers": {
"dominiclachance-neurakeep": {
"command": "npx",
"args": [
"-y",
"neurakeep"
]
}
}
}
Run commandnpx -y neurakeep

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages
npmneurakeep0.1.0

Tools

No tool list yet

Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.

Details

From the project README

NeuraKeep

NeuraKeep is a local-first memory system for AI agents.

It gives Claude Desktop, Claude Code, Codex, OpenClaw, Cursor-style tools, and other MCP clients a shared, source-cited memory layer that runs on your own machine.

What It Does

  • Imports Markdown, text, JSONL, and PDFs into a local vault
  • Preserves raw sources and extracts cited memory proposals
  • Lets you review memory before it becomes durable
  • Searches across sources, facts, events, failures, and decisions
  • Organizes the local app into grouped Explore, Workflow, and System navigation
  • Shows a Sigma knowledge graph with progressive expansion, clustering, search, keyboard access, and saved layouts
  • Exposes MCP tools for agents:
    • memory_search
    • memory_read
    • memory_explain_sources
    • memory_save (stores new text immediately and queues structured memory for review)
    • memory_propose_update
    • memory_apply_update

The core rule is simple: agents can save new source memory immediately, but extracted durable facts remain reviewable and source-cited.

Local Install

Requirements:

  • Node.js 22+
  • npm
npm install
npm run build:all
node dist/cli/index.js init --root .umos
node dist/cli/index.js serve --root .umos --host 127.0.0.1 --port 8765 --ui --read-write

Open:

http://127.0.0.1:8765/index.html

Start Free Locally

Use the local core when you want private, source-cited memory on your own machine:

  • import Markdown, text, JSONL, or PDFs
  • review extraction proposals before durable memory is applied
  • connect Claude, Codex, OpenClaw, or another MCP client over stdio

Hosted NeuraKeep

Use the hosted app when you need Google or enterprise SAML/OIDC sign-in, private personal vaults, isolated shared Team spaces, remote MCP over HTTPS, managed storage, backups, and Stripe-managed billing. Team workspaces include owner/admin/reviewer/editor/viewer roles, email-bound invitations, audit history, licensed seats, and SCIM provisioning. Hosted access requires an active individual or Team subscription. OAuth-capable remote MCP clients use browser sign-in and explicit space consent with PKCE and rotating refresh tokens; space-bound, scoped, expiring manual grants remain available for non-interactive clients. Hosted autonomous spaces use application-level encryption at rest, but are not zero-knowledge because the authorized runtime decrypts them while working.

The public package contains the complete local-first core. Managed authentication, Stripe billing, Team tenancy, SSO, SCIM, deployment automation, and production credentials remain part of the hosted service and are not required to run NeuraKeep locally.

CLI Examples

node dist/cli/index.js ingest README.md --root .umos --kind markdown --space personal --trust-level trusted --origin manual --sensitivity internal
node dist/cli/index.js search "what should my agents remember?" --root .umos --space personal --neighbors 1
node dist/cli/index.js propose latest --root .umos
node dist/cli/index.js apply <proposal-id> --root .umos
node dist/cli/index.js mcp --root .umos

Local MCP Setup

Example stdio MCP config:

{
  "command": "node",
  "args": [
    "C:/path/to/neurakeep/dist/cli/index.js",
    "mcp",
    "--root",
    "C:/path/to/neurakeep/.umos"
  ]
}

Local HTTP mode:

node dist/cli/index.js serve --root .umos --host 127.0.0.1 --port 8765 --ui

Use --token or NEURAKEEP_MCP_TOKEN before exposing any HTTP endpoint outside your machine.

Free vs Paid

The open-source version is for local, self-managed memory. The hosted product adds managed auth, team spaces, remote connectors, shared storage, backups, and administration. See NeuraKeep pricing for the current hosted plans.

Docs

Development Checks

npm run typecheck
npm run build:all
npm test
npm run e2e
npm run check:secrets

License

Apache-2.0

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.