Local-first persistent memory for AI coding agents, storing decisions and lessons in SQLite and serving them over MCP to Claude Code, Cursor and other clients.
arben-adm/mcp-sequential-thinking
A local MCP server that stores working notes for hard tasks in SQLite, so evidence, decisions and next steps can be resumed after a restart.
- Stars
- 959
- Forks
- 121
- Open issues
- 1
- Last commit
- 17 hours ago
What is it?
MCP Sequential Thinking is a local MCP server for keeping working notes on a difficult task. Each session holds notes of several kinds, such as evidence, assumptions, options, risks, decisions and next actions. Everything is stored durably in SQLite, so the notes can be read back after any restart.
The server records what the calling model or user writes. It does not generate reasoning, judge the content, or ask for hidden chain-of-thought. Reads come in a resume view that shows active notes with decisions first, or a full history view with filters. Notes can supersede earlier ones, branch into named lines, and reference a parent step, all within one session.
Writes take a request ID so retries are safe, and finalize and delete calls use an expected version to avoid conflicts. Sessions can be exported to and imported from JSON archives, and a backup tool snapshots all sessions. Storage is plaintext SQLite on a local disk, and it works with clients such as Claude Desktop and Cursor over stdio.
Key features
- 01Stores notes by session in local SQLite that survives restarts.
- 02Supports note kinds such as evidence, decision, risk and next_action.
- 03Resume view shows active notes with decisions first.
- 04Idempotent writes via request_id and version checks on finalize and delete.
- 05Exports and imports single sessions as JSON archives.
- 06Includes a backup and restore command for all sessions.
Use cases
- Keeping track of evidence and decisions across a long debugging or research task.
- Resuming an interrupted AI-assisted task in a new session without re-explaining context.
- Giving an MCP-capable assistant a durable scratchpad for risks and next actions.
Quick start
- 1
Create and activate a Python virtual environment
python -m venv .venv && . .venv/bin/activate - 2
Install the package in editable mode
python -m pip install -e . - 3
Run the restart demo to write notes, restart, and read them back
python scripts/stdio_smoke.py - 4
Add the server to your MCP client configuration (Claude Desktop, Cursor)
Things to know
- Session data is stored as plaintext SQLite, and deletion is not secure erasure.
- Storage must be on a local disk, since NFS, SMB and multiple hosts are not supported.
- HTTP and SSE transports are experimental.
- The legacy tools are deprecated and kept only for existing clients.
Topics
FAQ
- What is mcp-sequential-thinking?
- A local MCP server that stores working notes for hard tasks in SQLite, so evidence, decisions and next steps can be resumed after a restart. MCP Sequential Thinking is a local MCP server for keeping working notes on a difficult task. Each session holds notes of several kinds, such as evidence, assumptions, options, risks, decisions and next actions. Everything is stored durably in SQLite, so the notes can be read back after any restart.
- Is mcp-sequential-thinking free to use? What license does it have?
- mcp-sequential-thinking is open source under the MIT license. Check the license terms before using it in commercial products.
- Is mcp-sequential-thinking still maintained?
- Actively maintained. The last commit to mcp-sequential-thinking was on Oct 21, 2026.
- What can I use mcp-sequential-thinking for?
- Keeping track of evidence and decisions across a long debugging or research task. Resuming an interrupted AI-assisted task in a new session without re-explaining context. Giving an MCP-capable assistant a durable scratchpad for risks and next actions.
- How do I install mcp-sequential-thinking?
- Create and activate a Python virtual environment: python -m venv .venv && . .venv/bin/activate Install the package in editable mode: python -m pip install -e . Run the restart demo to write notes, restart, and read them back: python scripts/stdio_smoke.py Add the server to your MCP client configuration (Claude Desktop, Cursor)
- Where can I get mcp-sequential-thinking?
- The source code and installation instructions are on GitHub: https://github.com/arben-adm/mcp-sequential-thinking
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
A local-first personal AI assistant in readable Python, with its own agent loop, SQLite-based memory, and built-in evals.
Self-hosted, local-first personal AI agent with layered memory, tool permissions, workflows and IM channel integration.
An MCP server that lets AI assistants like Claude and GPT query and control devices on the ThingsPanel IoT platform and ThingsVis dashboards.
Local-first shared project memory layer that lets coding agents like Claude Code, Codex, and Cursor read and write the same memory via MCP.
Long-term memory for coding agents that stores memories in a local SQLite database and shares them across Claude Code, Codex, Cursor and other MCP clients.