Kith Mind
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Kith MindRemember that Sarah's new address is 12 Elm Street."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Kith Mind
Kith Mind is a personal knowledge base that an AI assistant can read and write. You tell Claude or ChatGPT something once. It is still there next week, in a new conversation, in the other assistant.
It stores three kinds of knowledge:
Kind | Example |
Facts | A provider name, a school, a policy number, a date |
Thoughts | A decision and why you made it, a project's state |
Documents | A statement or note you captured, with the text kept |
You host it yourself. The data sits in your own PostgreSQL database, not in a shared service. Assistants reach it over MCP, so no browser extension or copy-paste is involved.
Current account values come from a daily Plaid feed
(@repo/plaid-feed), not from parsing PDF statements; statements remain
monthly history and a cross-check, importable once as optional history older
than the feed. Documents come from a stateless, idempotent ingester
(@repo/ingest-simple) that walks a folder, hashes each file, and upserts by
hash -- no journal or durable worker to operate. Health records are planned
from Epic's patient-facing FHIR API, one authorization per family member; see
the simplification and feeds plan
for the full design.
Getting started
Setting up your own deployment, including the daily finance feed, the
document folder, and health records? docs/onboarding.md
is the full runbook. The steps below cover the core web app and MCP
connection only.
Related MCP server: Memsolus MCP Server
Use it with Claude or ChatGPT
Steps 1 through 3 happen once. Step 4 is the daily part.
Step 1. Deploy your own instance
Fork the repository, then follow
docs/self-hosting.md. It walks through one
PostgreSQL database, one Vercel project, and the environment variables that
connect them. Node.js 22 or newer and pnpm 10.20 are the only prerequisites.
You finish this step with an HTTPS address such as
https://your-project.vercel.app. Your MCP endpoint is that address plus
/api/mcp.
Expect a small monthly bill at most. A small hosted PostgreSQL instance and the Vercel free tier cover light personal use. Semantic search and narrative analysis call OpenAI and Anthropic and are metered, but both are optional. Keyword search and capture work without either key.
Step 2. Create your account
Open the deployed web app and sign up. This account owns your Personal space. Everything you capture goes there unless you later create a shared space. Password sign-in remains available. To add Google as a second sign-in method, follow the optional Google web sign-in setup, then connect the Google identity from Settings while signed in.
Step 3. Connect your assistant
In Claude, open Settings, then Connectors, then add a custom connector
pointing at https://your-project.vercel.app/api/mcp. Sign in through the
OAuth prompt using the account from step 2, then approve the permissions.
In ChatGPT, add the same URL as a custom MCP connector. Custom connectors require a paid plan.
In Claude Code, skip the browser:
claude mcp add --transport http kithmind https://your-project.vercel.app/api/mcpGrant read permission on your Personal space first. Add write and ingest permissions when you actually need them.
Step 4. Talk normally
There is no special syntax. Say durable things and the assistant stores them.
Our new pediatrician is Dr. Reyes at Lakeside Family Health.
We decided to keep the 2019 Outback instead of trading it in, mostly because the quote on the replacement was 9k over what we wanted to spend.
Ask about them later, in any connected assistant, in a fresh conversation:
Who is the kids' pediatrician?
Why did we keep the Outback?
The assistant decides when to call the tools. Kith Mind describes very clearly when each tool should be used, but it cannot see your conversation unless the client calls a tool. If an assistant is quiet about it, say "remember this" or "check my knowledge base" and it will.
Step 5. Confirm it actually stuck
Worth doing once, with a fact you do not mind being wrong:
State a fact, then ask for it back using different words.
Change the fact. Confirm you get the new value, and that asking what it used to be returns the old one as superseded.
Correct a fact you got wrong. It is retracted, not recorded as once true.
Repeat a fact. It does not duplicate.
Step 6. Add family, optionally
Use Spaces in the web app to create a shared space and invite someone. They accept an invite link, you approve them, and you both get shared access with Personal spaces still private. Roles, person links, and scoped API keys are in the family spaces guide.
What the assistant gets
Tool | Purpose |
| Precise facts, with history |
| Narrative memory and grounded recall |
| Captured documents and retained text |
| Typed lab, vehicle, and money records |
Set MCP_TOOL_PROFILE=memory to expose the memory and document tools only.
The default is the full set, which is currently the same set.
Where it stands
Working today: typed facts with supersession and retraction, narrative thoughts
with hybrid retrieval and citations, authenticated text capture, document
search with semantic and keyword modes, family spaces with invitations and
roles, capability-scoped API keys, an OAuth MCP gateway, a daily Plaid finance
feed, and a stateless document ingester for a local or provider-synced folder.
See docs/onboarding.md to set these up.
Not yet: Epic FHIR health records and cloud monitoring. The architecture document and the simplification and feeds plan are the source of truth for what is planned.
Mobile is deliberately thin. Desktop is the primary workflow, and hosted MCP covers mobile clients that support it. There is no native app planned.
Running it locally
pnpm install
cp apps/web/.env.example apps/web/.env.local # then set KITH_DATABASE_URL and friends
pnpm devpnpm demo:brain runs a synthetic end-to-end check that calls no AI provider.
Its four environment inputs are documented in
docs/self-hosting.md.
Before opening a pull request:
pnpm lint
pnpm check-types
pnpm test:once
pnpm buildCONTRIBUTING.md covers the contribution workflow and AGENTS.md covers agent guidance.
License
Kith Mind contributions are intended to be MIT-licensed. The repository derives from earlier work whose attribution is preserved in LICENSE and explained in provenance.
This server cannot be deployed
Maintenance
Related MCP Connectors
Persistent memory for AI agents. Search and store durable facts, preferences and decisions.
Universal persistent memory and knowledge retrieval layer for AI agents and LLMs.
Personal memory layer: files, notes, messages, tasks and email, semantically indexed.
Persistent memory for AI agents. Semantic search, memory graph, W3C DID identity.
Related MCP Servers
- FlicenseBqualityNot gradedmaintenanceProvides persistent personal and organizational knowledge storage that any LLM can connect to for instant context about who you are, what you do, and your preferences. Enables AI agents to maintain long-term memory across sessions through structured data categories, semantic search, and export/import capabilities.27-

Memsolus MCP Serverofficial
AlicenseAqualityDmaintenanceProvides persistent long-term memory for AI agents through semantic search and automated knowledge graph extraction. It enables agents to store, recall, and reason over facts, preferences, and relationships across multiple conversations and sessions.148 npmMIT- FlicenseNot gradedqualityDmaintenanceMCP server providing persistent memory for LLMs via semantic search over captures, structured facts with temporal tracking, and document ingestion. Enables any MCP-compatible client to maintain a searchable knowledge base across sessions.-
- AlicenseBqualityBmaintenanceProvides LLM clients with a persistent, scalable knowledge graph memory system that supports semantic retrieval, contextual recall, and temporal awareness.2128 npm1MIT