@jussmor/commit-memory-mcp

A local MCP server for PR intelligence, author tracing, and worktree-aware async planning.

Purpose

This package helps agents answer:

• who changed a file or area
• why a change was made
• what landed on main recently
• what context should be reviewed before planning

Current MCP tools

• sync_pr_context
• build_context_pack
• promote_context_facts
• archive_feature_context
• list_active_worktrees
• who_changed_this
• why_was_this_changed
• get_main_branch_overnight_brief
• resume_feature_session_brief
• pre_plan_sync_brief

Removed tools

The following legacy tools were removed:

• search_related_commits
• explain_commit_match
• get_commit_diff
• reindex_commits

Quick start

npm install @jussmor/commit-memory-mcp
npx commit-memory-mcp

For local development in this repository:

cd packages/commit-rag-mcp
npm install
npm run build
node dist/mcp/server.js

Requirements

• Node.js 20+
• gh CLI authenticated for GitHub PR sync features
• a git repository available through COMMIT_RAG_REPO

Environment variables

• COMMIT_RAG_REPO repository path used by the MCP server
• COMMIT_RAG_DB SQLite database path
• COMMIT_RAG_LIMIT default sync/query limit
• OLLAMA_BASE_URL optional Ollama base URL
• OLLAMA_EMBED_MODEL optional embedding model
• COPILOT_TOKEN optional Copilot reranking token
• COPILOT_MODEL optional Copilot model override
• COPILOT_BASE_URL optional Copilot API base URL

Use cases

Sync GitHub PR context

sync_pr_context({
  owner: "MaxwellClinic-Development",
  repo: "EverBetter-Pro",
  domain: "billing",
  feature: "invoice-retry",
  branch: "feat/invoice-retry",
  taskType: "planning",
  limit: 20
})

Use this before planning when you need fresh PR descriptions, comments, and reviews.

List active worktrees

list_active_worktrees({
  baseBranch: "main"
})

Use this when your team works on multiple features in parallel and wants session-aware context.

Build a scoped context pack

build_context_pack({
  domain: "billing",
  feature: "invoice-retry",
  branch: "feat/invoice-retry",
  taskType: "coding",
  includeDraft: false,
  limit: 12
})

Use this before invoking a coding subagent to keep prompts small and focused. The pack is assembled main-first: promoted facts on main are treated as baseline truth, then branch/feature facts are added as in-flight overlay. If no rows are found in strict scope, the server falls back automatically to broader scope levels.

Important: if you provide domain/feature/branch tags in build_context_pack, use the same tags during sync_pr_context for best precision.

Promote draft facts after review

promote_context_facts({
  domain: "billing",
  feature: "invoice-retry",
  branch: "feat/invoice-retry"
})

Use this when discussion outcomes are approved and should become durable context.

Archive completed feature context

archive_feature_context({
  domain: "billing",
  feature: "invoice-retry"
})

Use this after merge/closure to prevent active context bloat.

Find ownership for a file

who_changed_this({
  filePath: "src/features/auth/session.ts",
  limit: 15
})

Use this to discover recent authors and commit history for a target file.

Explain what is happening in a Next.js folder

explain_path_activity({
  targetPath: "app/dashboard",
  owner: "MaxwellClinic-Development",
  repo: "EverBetter-Pro",
  limit: 30
})

Use this when you want a fast folder-level summary for areas like app/dashboard, app/(authenticated), or src/components. It returns recent commits, top authors, most touched files, and related PR context/decisions when available.

Explain intent for a change

why_was_this_changed({
  filePath: "src/features/auth/session.ts",
  owner: "MaxwellClinic-Development",
  repo: "EverBetter-Pro"
})

Use this to combine commit metadata with synced PR context.

Get an overnight main-branch brief

get_main_branch_overnight_brief({
  baseBranch: "main",
  sinceHours: 12,
  limit: 25
})

Use this at the start of the day to review what landed while you were offline.

Resume a feature worktree

resume_feature_session_brief({
  worktreePath: "/path/to/worktree",
  baseBranch: "main"
})

Use this before resuming unfinished work in a separate worktree.

Run the full pre-plan sync flow

pre_plan_sync_brief({
  owner: "MaxwellClinic-Development",
  repo: "EverBetter-Pro",
  baseBranch: "main",
  worktreePath: "/path/to/worktree",
  sinceHours: 12,
  limit: 25
})

Use this as the default entrypoint for async team planning.

Multi-session git worktree flow

For parallel AI coding sessions:

1. Create one git worktree per feature branch.
2. Use list_active_worktrees to enumerate active sessions.
3. Use resume_feature_session_brief per worktree to check divergence and overlap risks.
4. Generate a worktree-specific build_context_pack and hand it to the target subagent.

This pattern avoids one giant shared context and scales better as features grow.

Data model overview

The package stores local context for:

• commits
• pull requests
• PR comments
• PR reviews
• promoted decision/blocker summaries
• worktree session checkpoints

Publishing

npm run build
npm publish --access public

For MCP Registry publication, keep package.json mcpName and server.json name aligned.