cc-dash
A local dashboard and Model Context Protocol server for managing a multi-project portfolio whose entire state lives in markdown files.
README
cc-dash
A local dashboard and Model Context Protocol server for managing a multi-project portfolio whose entire state lives in markdown files.
Projects declare their state in ROADMAP.md, SESSION_PROGRESS.md, and QA.md files
using a small schema. cc-dash reads those files, renders them as boards, timelines, and
queues, and writes changes back to the same markdown. There is no database. The
dashboard is a lens over your files, not a copy of them.
The MCP server exposes the same operations to Claude Code, so an agent can read a roadmap, advance a session, or work a QA queue without anyone opening the UI.
The MCP server
src/mcp/server.ts is a standard MCP server over stdio, with 25 zod-typed tools.
claude mcp add --scope user cc-dash -- npx tsx /path/to/cc-dash/src/mcp/server.ts
Tools are grouped by what they act on:
| Group | Tools |
|---|---|
| Portfolio | list_projects, get_portfolio, get_project, get_config, search, reorder_projects |
| Roadmap | add_roadmap_item, add_roadmap_category, update_roadmap_item, bulk_update_status |
| Sessions | get_session, list_sessions_today, update_session_status |
| QA | next_qa_item, approve_qa_item, fail_qa_item, skip_qa_item, reset_qa_item |
| Planning | get_today_directions, list_ideas, set_project_status |
Every tool validates its input with zod and returns structured content. A tool that cannot do what was asked raises instead of returning a plausible-looking empty result, so an agent cannot silently proceed on bad state.
Why markdown as the database
Three properties fall out of that constraint, and they are the reason it is worth its cost:
- The files stay useful without the app. A
ROADMAP.mdis readable in any editor, diffable in any review, greppable from any shell. - Git is the audit log. Every state change is a commit: attributable, revertible, and with no separate migration story.
- Agents already understand it. A coding agent can read and edit the source of truth directly, with or without the MCP server.
The cost is that parsing and serialization have to be exact. Round-trip fidelity is enforced by tests: parse a file, serialize it, and the bytes must match.
Features
- Project discovery across configured scan directories
- Roadmap board with kanban and list views, drag and drop, and bulk status edits
- Session tracking for tasks, decisions, and failed attempts across work sessions
- QA queues, portfolio-wide and per-project, with a keyboard-driven focus mode
- Today's Directions, a generated daily plan with dispatch commands per project
- Agent activity feed that attributes commits to the AI agents that co-authored them
- Ideas system with a guided capture wizard
- Milestone timeline built from roadmap and session events
Setup
Requires Node.js 22 or newer.
npm install
npm run dev # http://localhost:3000
npm run mcp # run the MCP server over stdio
Create ~/.config/cc-dash/config.json:
{
"scan_dirs": ["~/projects"],
"exclude_dirs": ["node_modules", ".git", "vendor"],
"explicit_projects": [],
"scan_depth": 2,
"port": 3000,
"orchestrator_dir": "~/projects"
}
orchestrator_dir is the portfolio-level directory your orchestrator agent runs from,
and where TODAYS_DIRECTIONS.md is written. Every machine-specific path lives in this
file, outside the repository, so nothing personal is ever committed.
A project is discovered when it contains a schema file:
---
schema: cc-dash/roadmap@1
project: my-project
description: One-line description
last_updated: 2026-01-01T00:00:00Z
---
Roadmap items carry stable IDs in HTML comments, so an item can be reworded without losing its identity:
- <!-- id:r_a1b2c status:planned --> **Item name** - What it does
Commands
| Command | Description |
|---|---|
npm run dev |
Dev server |
npm run build |
Production build |
npm run mcp |
MCP server over stdio |
npm run test:run |
Full suite, once |
npm test |
Watch mode |
npm run test:e2e |
Playwright end-to-end |
npm run lint |
ESLint |
1,666 tests across 148 files.
Constraints
These are enforced, not aspirational:
- No database. Markdown files are the only state.
- Round-trip fidelity. Parse then serialize must reproduce the file byte for byte.
- Atomic writes. Every mutation writes to a
.tmpfile, then renames. - Item IDs are permanent. Once assigned, an ID never changes.
Stack
Next.js 15, React 19, TypeScript, Tailwind CSS 4, zod,
@modelcontextprotocol/sdk,
Vitest, Playwright.
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。