MCP Server: Context & Todo Persistence with AI

Overview

MCP Server (Model Context Protocol) is a lightweight service for saving, restoring, and sharing development context: code snippets, task lists (todos), and arbitrary metadata that make it easier to pick up work or hand off context between sessions and teammates. It also integrates with multiple large language model providers so you can ask an LLM for a second opinion on code, design decisions, or a todo list while preserving the surrounding context.

The server is useful when you want to persist the state of a coding session beyond your editor: snapshot code fragments, attach related todos, and ask ChatGPT, Claude, Gemini, or DeepSeek to review or summarize that snapshot. Compared with ephemeral chat windows, an MCP server helps keep context durable, searchable, and shareable across tools and team members.

Features

• Save and restore developer context: titles, descriptions, metadata
• Persist code snippets and associated files
• Store and manage todo items linked to a context
• Query multiple LLM providers for “second opinions” (ChatGPT, Claude, Gemini, DeepSeek)
• Multi-provider adapter architecture — swap or configure different LLM backends
• Simple HTTP API and optional web UI for browsing contexts and snippets
• Authentication and API key configuration for external LLM services
• Designed to be self-hosted (Docker / local) and lightweight to run alongside dev tooling

Installation / Configuration

Clone the repository and choose a run mode (Docker or local). The project includes configurations to run as a container or as a local service.

Install with Git and Docker (recommended)

git clone https://github.com/taylorleese/mcp-toolz.git
cd mcp-toolz
docker-compose up -d

Run locally (example)

git clone https://github.com/taylorleese/mcp-toolz.git
cd mcp-toolz
# If the repo is a Node/Python app, install dependencies (example shown generically)
npm install        # or: pip install -r requirements.txt
npm run start      # or: uvicorn app.main:app --reload

Environment variables Create a .env file at the project root with the keys your deployment needs. Typical variables:

# Storage / server
MCP_PORT=8080
DATABASE_URL=postgresql://user:pass@localhost:5432/mcpdb

# Authentication (app-level secret)
MCP_SECRET_KEY=supersecret

# LLM provider keys (set whichever providers you use)
OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=anthropic-...
GOOGLE_API_KEY=ya29....   # for Gemini
DEEPSEEK_API_KEY=ds-...

Table: common env variables

Check the repo README for extra configuration flags (logging, persistence backends, CORS).

Available Resources

• Source code and issues: https://github.com/taylorleese/mcp-toolz
• HTTP API: endpoints to create/read/update/delete contexts, snippets, and todos (see API docs in repo)
• LLM adapter layer: configure one or more providers and switch per-query
• Web UI (if bundled): browse saved contexts, open associated snippets, trigger AI reviews

Example API resources (conceptual)

• POST /api/contexts — create a new context snapshot
• GET /api/contexts/{id} — load a saved context and its snippets/todos
• POST /api/contexts/{id}/ask — query an LLM for a review or summary of the context

Use Cases

• Session snapshots: Save a “workspace” before switching tasks. Persist code snippets, command outputs, and todos so you can return later with full context.

  • Example: Save a context named “bugfix/X” with a failing test snippet and the reproduction steps; later ask ChatGPT for debugging hints.

• PR / code review assist: Attach the relevant snippets and design notes to a context and request a second opinion or summary from an LLM to include in a review comment.

  • Example: POST a context with the new function and ask Claude to summarize complexity and edge-cases.

• Onboarding and handoff: Team members can save step-by-step notes and code fragments about a feature. New contributors pull the context and query Gemini for an overview.

  • Example: Create a context “feature/metrics” with todos and a short explainer, then send link to a new teammate.

• Persistent todos tied to code: Keep TODOs next to the snippet they refer to so you don’t lose track when switching machines or editors.

  • Example: Save a snippet with inline todos for refactoring tasks; retrieve them later and mark complete via the API.

• Comparing LLM opinions: Query multiple providers on the same context to compare suggestions or risk assessments.

  • Example: POST /api/contexts/{id}/ask with provider=chatgpt and provider=claude and compare outputs.

Getting Help

• Open an issue on the project repo: https://github.com/taylorleese/mcp-toolz/issues
• Inspect the repository README and API docs for endpoint details and sample payloads
• If running in production, secure your MCP_SECRET_KEY and limit access to the API to trusted networks

This server is intended to be a practical scaffold for persistent developer context and multi-LLM integrations — useful for personal workflows, team collaboration, and building tooling that needs durable, shareable development state.