Labsco
MCP SERVER

MCP Conductor

by Lutherscottgarcia ↗

Cache what an assistant learned about your project once, and load it back at the start of the next session.

Agent Memory
Summary
A cache is only worth loading if you know it is still true.

The interesting part is not the write, it is `validate_project_intelligence_cache` and `invalidate_project_cache` — a stale picture of a codebase is worse than no picture, because the assistant acts confidently on it. Treat the validate call as part of the load ritual rather than an optional extra. Note the platform line before you commit: this is built and supported on macOS, with Linux flagged experimental.

What it is

A project-context cache for Claude Desktop. It captures an understanding of a codebase — architecture, technologies, momentum, next steps — into a Memory MCP entry, so a later session loads that entry instead of re-exploring the repository. Around it sits a session-rules engine and coordination across Memory, Filesystem, Git, Claudepoint and database MCPs.

What you get
  • `create_project_intelligence_cache` writes the current understanding of a project into Memory MCP
  • `load_project_intelligence_cache` reads it back at the start of a session
  • `validate_project_intelligence_cache` checks whether the cached picture still matches the project
  • `refresh_project_intelligence` updates a cache after the project has moved on
  • `invalidate_project_cache` drops one when it can no longer be trusted
  • A session-rules engine with enforcement levels from `hard_block` down to `log_only`, so a rule can stop an action or merely record it
  • Project directories are auto-detected across the common layouts — `~/Projects/`, `~/Developer/`, `~/Code/`, `~/RiderProjects/` and others
Requirements

Node.js 18.0.0 or higher and npm 8.0.0 or higher, plus Claude Desktop. Published as conversation-continuity-mcp version 1.0.0. Two MCP dependencies are installed for you: `@modelcontextprotocol/server-memory` and `@modelcontextprotocol/server-filesystem`. It needs read and write access to your project directories and to `~/Library/Application Support/Claude/`. macOS 10.15+ is the supported platform; Linux is marked experimental and Windows is not there yet. `MCP_CONDUCTOR_PROJECT_DIR` and `MCP_CONDUCTOR_WORKSPACE` override the auto-detected paths.

Setup effort

One command — curl -fsSL https://raw.githubusercontent.com/lutherscottgarcia/mcp-conductor/main/install.sh | bash