# Home Lab Context MCP

A small, storage-backed Model Context Protocol server for sharing durable project
context between trusted coding agents. It is designed for a private network or
another controlled environment; it is not a public data-sharing service.

## What it provides

- Project and safe-text-file discovery.
- Read operations with ETags for optimistic concurrency.
- Writes limited to `context/` and `handoffs/`.
- Timestamped handoffs for agent-to-agent continuity.
- No delete operation.

## Shared agent memory

```mermaid
flowchart LR
  Claude["Claude Code"] <--> MCP["Home Lab Context MCP"]
  Codex["Codex"] <--> MCP
  MCP <--> Store["S3-compatible context store"]
  Store --> State["Current state"]
  Store --> Handoff["Decisions, tasks, handoffs"]
```

The MCP is deliberately shared memory. Private chat history remains with each
agent; only explicit project context is shared.

## Security model

The server deliberately rejects path traversal, secret-like names, unsupported
file types, and oversized requests. Deploy it behind an authenticated ingress,
use least-privilege storage credentials, and never store secrets in shared
context. See [the threat model](docs/threat-model.md).

## Repository status

This repository contains the reusable authenticated HTTP MCP server, its
S3-compatible context store, optimistic concurrency, and timestamped handoffs.
It contains no private endpoint, credential, network, or cluster configuration.

## Development

```bash
python -m unittest discover -s tests
python server.py
```

## License

Apache-2.0. See [LICENSE](LICENSE).