Local Context MCP

AI agents forget everything the moment a session ends. This is a small, self-hosted memory service that fixes that: agents and tools write durable facts to it, and read them back later — across sessions, across machines, across whichever client you happen to be using that day.

The problem

Every AI coding tool re-derives the same context from scratch: your conventions, your decisions, your gotchas. You end up repeating yourself in every new chat, or copy-pasting the same paragraph into every CLAUDE.md. Memory should be a shared, queryable fact store — not something re-explained by hand each time.

The approach

• One project, one door. REST API, MCP server, and dashboard are three interfaces on the same Laravel application — not three services to keep in sync.

• The project key is the whole security model. Every memory belongs to a project, identified by a bearer key. REST, MCP, and the dashboard all resolve tenancy the same way, so there's exactly one mental model to hold.

• No new infrastructure. Memories are searched with MySQL FULLTEXT, not a vector database — this is built for roughly a dozen projects with hundreds of memories each, and boring technology is enough at that scale. Revisit if that assumption changes.

• Deploy once, use everywhere. The server can run on your laptop or on a VPS; any MCP-compatible client (Claude Code, Devin, Cursor, ...) can register against it without needing this repository checked out locally.

flowchart LR
    A[Claude Code] -->|remember / recall| S[(Local Context MCP)]
    B[Devin] -->|remember / recall| S
    C[REST client] -->|remember / recall| S
    S --> D[(MariaDB/MySQL)]

Get started

If you already have a project key from a server someone else deployed, you don't need any of the setup steps or even this repository — go straight to Registering MCP clients, which includes a one-line install script.