@mempalace/mcp-dev
Enables AI coding assistants to access grounded, branch-scoped codebase context via semantic search, git tracking, change ledger, and structured feature management with Project Tracks.
README
@mempalace/mcp-dev
An MCP (Model Context Protocol) server that gives AI coding assistants grounded, branch-scoped context about your codebase. It provides semantic code search via my MemPalace fork, git change tracking, a persistent change ledger, TypeScript/Jest dev lifecycle tools, and a built-in Project Tracks (Conductor) methodology for structured feature work.
Features
| Feature | Description |
|---|---|
| Semantic code search | Natural-language queries over your indexed codebase via MemPalace embeddings |
| Branch context | Current branch, SHA, recent commits — always oriented |
| Change ledger | Append-only JSONL log of every AI-assisted change with full provenance |
| Dev lifecycle | Run tsc typecheck and Jest tests from within the MCP client |
| Git diff | Full or file-scoped diff output |
| Project Tracks | Create and manage markdown-based feature tracks (plan/spec/index/metadata) |
Prerequisites
- Node.js ≥ 18
- Python ≥ 3.9
- A git repository to point at
Important: MemPalace Fork Required
⚠️ This MCP server requires a custom fork of MemPalace with multi-branch exclude-patterns support. The official
pip install mempalacepackage will not work — you must use my fork (branchfeat/exclude-patterns-config).
The easiest way to get this set up is the setup script (see Quick Start).
Quick Start
1. Run the setup script (recommended)
The fastest way to get up and running is the bootstrap script at
scripts/setup-mempalace.sh. It handles the entire MemPalace setup:
- Clones my fork with the required exclude-patterns support
- Creates a Python virtual environment
- Installs the fork in editable mode
- Symlinks the
mempalaceCLI to~/.local/bin - Prepares the palace data directory
- Optionally runs the initial codebase index
# From the repo root:
./scripts/setup-mempalace.sh
# To also index your codebase right away:
./scripts/setup-mempalace.sh --mine
The script is fully configurable via environment variables:
| Env var | Default | Description |
|---|---|---|
MEMPALACE_FORK_URL |
https://github.com/2loch-ness6/mempalace |
My fork with required exclude-patterns support |
MEMPALACE_FORK_BRANCH |
feat/exclude-patterns-config |
Branch containing the feature |
MEMPALACE_FORK_DIR |
$HOME/.mempalace-fork |
Where to clone the fork |
MEMPALACE_VENV_DIR |
$HOME/.mempalace |
Python venv location |
MCP_PALACE_DIR |
<repo-root>/.palace/active |
Palace data directory |
MCP_PALACE_WING |
code |
Palace wing (namespace) |
MCP_REPO_DIR |
<repo-root> |
Source code to mine |
After the script finishes, check that everything works:
mempalace --version
mempalace --palace ~/.mempalace/active status
2. Install the MCP server
npm install -g @mempalace/mcp-dev
Or use without installing via npx:
npx @mempalace/mcp-dev
3. Configure environment variables
export MCP_REPO_DIR=/path/to/your/repo # defaults to process.cwd()
export MCP_PALACE_DIR=~/.mempalace/active # MemPalace data directory
export MCP_PALACE_WING=code # wing name (namespace) for your repo
export MEMPALACE_PYTHON=/path/to/venv/bin/python3 # python with fork installed
If you used the setup script, set
MEMPALACE_PYTHONto$HOME/.mempalace/bin/python3(or whereverMEMPALACE_VENV_DIRpoints).
4. Add to your MCP client config
For Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"mempalace-mcp-dev": {
"command": "npx",
"args": ["@mempalace/mcp-dev"],
"env": {
"MCP_REPO_DIR": "/path/to/your/repo",
"MCP_PALACE_DIR": "/home/you/.mempalace/active",
"MCP_PALACE_WING": "code",
"MEMPALACE_PYTHON": "/home/you/.mempalace/bin/python3"
}
}
}
}
For Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"mempalace-mcp-dev": {
"command": "npx",
"args": ["@mempalace/mcp-dev"],
"env": {
"MCP_REPO_DIR": "/path/to/your/repo",
"MCP_PALACE_DIR": "/home/you/.mempalace/active",
"MCP_PALACE_WING": "code",
"MEMPALACE_PYTHON": "python3"
}
}
}
}
Environment Variables
| Variable | Default | Description |
|---|---|---|
MCP_REPO_DIR |
process.cwd() |
Absolute path to your git repository root |
MCP_PALACE_DIR |
~/.mempalace/active |
MemPalace data directory |
MCP_PALACE_WING |
code |
Wing (namespace) name within the palace |
MCP_LEDGER_PATH |
<repo>/.mcp-dev/ledger.jsonl |
Path to the change ledger file |
MCP_TRACKS_DIR |
<repo>/conductor/tracks |
Directory containing project track folders |
MEMPALACE_PYTHON |
python3 |
Python binary (must have the MemPalace fork installed) |
MCP Tools Reference
Code Search
| Tool | Description |
|---|---|
search_code |
Semantic search over indexed codebase. Accepts query (string) and optional scope (room name) and limit (int, default 10) |
mine_changed_files |
Re-index files changed since the last git commit. Call after editing to keep search current |
Branch & Git
| Tool | Description |
|---|---|
get_branch_context |
Returns current branch, SHA, short SHA, and last 5 commit messages |
get_changed_files |
Files changed vs HEAD~1 (or a specified base ref) with add/modify/delete status |
get_git_diff |
Full diff output. Optional base ref and filePath for scoping |
Change Ledger
| Tool | Description |
|---|---|
record_change |
Append an entry to the ledger (files affected, description, reason, tags, optional track link) |
read_ledger |
Read recent ledger entries. Optional limit (default 20) and branch filter |
Dev Lifecycle
| Tool | Description |
|---|---|
run_typecheck |
Run tsc --noEmit in a service directory. Returns structured errors with file/line/col/code |
run_tests |
Run Jest in a service directory. Returns pass/fail counts and structured failure details |
Project Tracks (Conductor)
| Tool | Description |
|---|---|
list_tracks |
List all live tracks (directories containing plan.md) |
get_track_plan |
Read plan.md for a track |
update_track_plan |
Overwrite plan.md for a track |
get_track_spec |
Read spec.md for a track |
update_track_spec |
Overwrite spec.md for a track |
get_track_index |
Read index.md for a track |
update_track_index |
Overwrite index.md for a track |
create_track |
Scaffold a new track directory with all four files |
Project Tracks Methodology (Conductor)
Project Tracks is a lightweight, file-based feature management system. Each track is a directory containing up to four files:
conductor/tracks/
└── my-feature-20260501/
├── plan.md ← living task list; the status source of truth
├── spec.md ← technical spec, design decisions, acceptance criteria
├── index.md ← short summary and linked resources
└── metadata.json ← machine-readable metadata (id, owner, status, dates)
Rules:
- A directory is a live track only if it contains
plan.md. plan.mdis always read/written live from disk (never via MemPalace — no staleness).- Track slugs must match
[a-zA-Z0-9][a-zA-Z0-9_-]*(max 100 chars).
Bootstrapping tracks in a new project:
# Create the tracks directory
mkdir -p conductor/tracks
# Use the MCP tool to scaffold a track
# (or copy from templates/track/ in this package)
Template files are included in templates/track/ for manual scaffolding.
MemPalace Configuration (mempalace.yaml)
Place a mempalace.yaml at the root of MCP_REPO_DIR to control how files are indexed. A generic example is provided in mempalace.yaml.example.
Key sections:
wing— must matchMCP_PALACE_WINGexclude_patterns— gitignore-syntax patterns to skip during miningrooms— semantic routing rules (keyword → room name) for search scoping
Development
git clone https://github.com/your-org/mempalace-mcp-dev
cd mempalace-mcp-dev
npm install
npm run build
npm test
See CONTRIBUTING.md for contribution guidelines.
License
MIT — see LICENSE.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。