CodeLedger solves the cold start problem for AI coding sessions by building persistent, deterministic context selection into your repo. It exposes three MCP tools: query_ledger to search verified patterns before writing code, get_active_context to pull the minimal file bundle for your current task, and record_interaction to capture outcomes that compound into institutional memory. Context persists across agents and sessions, so successful patterns from one task automatically inform the next. Works locally with no cloud dependency. The MCP server runs via stdio transport and requires Team or Enterprise tier, though the core context selection and CLI work in the free Individual tier.
claude mcp add --transport stdio codeledgerecf-codeledger uvx codeledger