Standalone MCP Server Packaging
Package the Pyrite MCP server as a standalone installable (`pip install pyrite-mcp`) so AI agent users can get a working MCP knowledge server in under 60 seconds without installing the full Pyrite CLI, web server, or extensions.
Motivation
This is the #1 go-to-market blocker for the agent shared memory market. Today, using Pyrite as an MCP server requires cloning the repo, installing all dependencies, and configuring a KB. The target experience is:
```bash
pip install pyrite-mcp
pyrite-mcp init my-research # creates KB + config
pyrite-mcp serve --tier write # starts MCP server
```
Scope
Extract `pyrite-mcp` as a minimal PyPI package containing:
- MCP server (`mcp_server.py`)
- Storage layer (database, repository, index)
- Services (kb_service, search_service)
- Schema and models
- Format serializers
Exclude: web server, CLI beyond `init`/`serve`, extensions, LLM service
Single `pyrite-mcp init` command that creates a KB directory with `kb.yaml`
`pyrite-mcp serve --tier read|write|admin` for stdio MCP transport
Publish to PyPI
Add to MCP community registryAcceptance Criteria
[ ] `pip install pyrite-mcp` works in a clean virtualenv
[ ] `pyrite-mcp init ` creates a working KB
[ ] `pyrite-mcp serve --tier write` starts an MCP server over stdio
[ ] All MCP tools, prompts, and resources work
[ ] Package size is minimal (no web frontend, no optional AI deps)
[ ] README with 3-minute quickstart for Claude Desktop / Claude CodeReferences
Agent shared memory positioning: `kb/positioning/agent-shared-memory.md`