Provides LLMs with a SQLite-backed project memory to track tasks, decisions, architecture notes, and session context for AI coding agents.
Also known as umitkavala/mindpm
- Stars
- 9
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- 3.2.0
- Last commit
- Oct 5, 2026
- License
- Apache-2.0
What is mindpm MCP?
mindpm provides LLMs with a SQLite-backed project memory to track tasks, decisions, architecture notes, and session context for AI coding agents.
Summary generated from the project's README
About mindpm MCP server
mindpm provides a persistent memory layer for AI coding agents by using a local SQLite database to store project state. Instead of relying on chat history, LLMs use this server to track tasks, architectural decisions, technical notes, and session context. This allows developers to resume work across different conversations without re-explaining project details or previous choices.
Users can integrate this server with clients like Claude Code, Cursor, VS Code, Cline, or Windsurf. Setup involves configuring the MCP client with a database path, such as ~/.mindpm/memory.db, and running the server via npx. The system includes a Kanban board accessible at localhost:3131 to manage work. It also features a mechanism to prevent parallel agents from colliding through claims and leases, and offers an optional verifier for unattended agent runs.
This tool is designed for developers who want to maintain a continuous project brain. It manages the lifecycle of tasks through tools like start_session and end_session, ensuring that every new session understands what was completed and what remains to be done.
What you can do with it
- 1Resume interrupted coding sessionsAsk the AI to identify the next priority task based on the previous session's progress.
- 2Coordinate multiple AI agentsPrevent different agents from working on the same task simultaneously using a leasing system.
- 3Automate unattended task verificationRun an automated check on commits submitted by agents that operate without human supervision.
Written from the project's README. Every use case is checked against a quote from it.
How to install mindpm MCP
Formats verified 10 Nov 2026Add to CursorAdd to GooseInstall mindpm MCP in Claude Code
$ claude mcp add mindpm -- npx -y mindpmInstall mindpm MCP in Cursor
~/.cursor/mcp.json
{
"mcpServers": {
"mindpm": {
"command": "npx",
"args": [
"-y",
"mindpm"
]
}
}
}Install mindpm MCP in VS Code
.vscode/mcp.json
{
"servers": {
"mindpm": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"mindpm"
]
}
}
}Install mindpm MCP in VS Code CLI
$ code --add-mcp '{"name":"mindpm","type":"stdio","command":"npx","args":["-y","mindpm"]}'Install mindpm MCP in Claude Desktop
claude_desktop_config.json
{
"mcpServers": {
"mindpm": {
"command": "npx",
"args": [
"-y",
"mindpm"
]
}
}
}Install mindpm MCP in Codex
$ codex mcp add mindpm -- npx -y mindpmInstall mindpm MCP in Codex config.toml
~/.codex/config.toml
[mcp_servers.mindpm] command = "npx" args = ["-y", "mindpm"]
Install mindpm MCP in Gemini CLI
$ gemini mcp add -s user mindpm npx -- -y mindpmInstall mindpm MCP in Gemini settings.json
~/.gemini/settings.json
{
"mcpServers": {
"mindpm": {
"command": "npx",
"args": [
"-y",
"mindpm"
]
}
}
}Configuration
MINDPM_DB_PATHPath to the SQLite database file
MINDPM_HOSTInterface the Kanban UI binds to. Loopback by default; set 0.0.0.0 only to opt in (e.g. under WSL2 NAT)
MINDPM_PORTPort for the built-in Kanban board UI
Automated checks
Ran Nov 14, 2026Full report →Passed our automated checks. That is not a security audit: read the code and give the server the narrowest access it needs. How checks work
- Source repository linked and reachableumitkavala/mindpm is public and reachablegithub.com/umitkavala/mindpm ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)No build attestation published for this releaseNot checked
- Open source license (OSI)Apache-2.0spdx.org/licenses/Apache-2.0.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-10-05github.com/umitkavala/mindpm ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNo popular package one edit away from mindpmPass
- No critical advisory in the package or its dependencies134 resolved dependencies checked; no critical advisoryosv.dev ↗Pass
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubmindpm
Project memory and a control plane for AI coding agents.
mindpm is an MCP (Model Context Protocol) server that gives LLMs a SQLite-backed brain for your projects. It tracks tasks, decisions, architecture notes, and session context — so every new conversation picks up exactly where you left off. It also hands specced tasks to coding agents, keeps parallel agents from colliding, and only counts work as done once a human has accepted it. For agents that run unattended, an optional verifier reruns the checks itself first.
The Problem
Every new LLM chat starts from zero:
- "Let me remind you about my project..."
- "Last time we decided to use Redis for..."
- "Where did we leave off?"
The Solution
mindpm persists your project state in a local SQLite database. The LLM reads and writes to it via MCP tools. No chat history needed. No memory features needed.
You: "What should I work on next?"
LLM: [queries mindpm] "Last session you finished the auth refactor.
You have 3 high-priority tasks: rate limiting, API docs, and
the webhook retry bug. Rate limiting is unblocked — start there."
What It Tracks
- Tasks, specs and acceptance criteria, with handoff statuses, priorities and blockers
- Decisions: what was decided, why, and what was rejected
- Notes and context: architecture, bugs, ideas, tech stack, conventions
- Sessions: what was done and what's next, plus a brief of what changed while you were away
- Agent attempts: every run on a task, and what the next attempt should avoid
Quick start
claude mcp add mindpm -e MINDPM_DB_PATH=~/.mindpm/memory.db -- npx -y mindpm
That's Claude Code, and npx fetches mindpm, so there's nothing to install first. Claude Desktop, Cursor, VS Code, Cline and Windsurf use the same JSON config in a different file. Then just talk about your project. The LLM calls start_session to load its context, records tasks, decisions and notes as you go, and calls end_session to save what's next. A Kanban board runs at http://localhost:3131.
How it works
┌──────────────┐ MCP ┌─────────┐ SQLite ┌───────────┐
│ Claude Code, │ ◄──────────► │ mindpm │ ◄────────────► │ memory.db │
│ Cursor, ... │ tools │ server │ read/write │ │
└──────────────┘ └─────────┘ └───────────┘
Everything stays on your machine, in one SQLite file (~/.mindpm/memory.db by default). mindpm can also hand specced tasks to coding agents with no conversation context: claims with leases keep parallel agents from colliding, and you accept submitted work on the board. For agents that run unattended, an optional verifier reruns the checks on each submitted commit first.
Security: mindpm has no user accounts and can't stop a process on your own machine. An agent with a shell can call the board's routes as you or edit the database directly. Read the security model before you let agents run unattended.
Documentation
Full documentation: https://umitkavala.github.io/mindpm/
- Setup: every MCP client, agent instructions for any LLM, storage
- Kanban board: lanes, accepting work, network binding and WSL
- Session brief: what changed while you were away
- Agent execution: lifecycle, specs, claims and briefs
- Advanced: unattended agents and verification
- Security model
- MCP tools
- Development
- Changelog
License
MIT
mindpm MCP questions
How do I add mindpm MCP to Claude Code?
Run claude mcp add mindpm -- npx -y mindpm in your terminal. Add --scope user to make it available in every project.
How do I add mindpm MCP to Cursor?
Open ~/.cursor/mcp.json and add the server under mcpServers, using the Cursor config shown in the install section above.
How does it prevent multiple agents from working on the same task?
It uses claims with leases to prevent parallel agents from colliding.