Owlyn MCP
Time-aware work sessions for AI agents.
README
<p align="center"> <img src="./Owlyn.png" alt="Owlyn MCP logo" width="760" /> </p>
Owlyn MCP
<p align="center"> <strong>Time-aware work sessions for AI agents.</strong> </p>
AI agents often stop too early.
You ask an agent to work for the night.<br /> It finishes the first task in 20 minutes.<br /> Then it says "done".
Owlyn gives agents a persistent work-session clock, checkpoints, deadlines, and structured continuation decisions so they know when to keep going.
Owlyn MCP is a local-first MCP work-session supervisor for AI coding agents. It is not a simple timer, a todo app, or a background worker.
Status: v0.1.0-alpha.1<br />
Tested with official MCP SDK stdio integration. Validated in Codex MCP host on Windows. Other hosts pending.
Quick Start
Requirements:
- Node.js 20+
- npm
Install and build:
npm install
npm run build
Run the stdio MCP server:
node dist/index.js
Run the release checks:
npm run check
Run the manual MCP smoke test:
npm run smoke:mcp
The smoke test uses the official MCP SDK client and calls owlyn_start, owlyn_checkpoint, owlyn_should_continue, owlyn_plan_next, and owlyn_end.
Example MCP Config
Generic stdio config:
{
"mcpServers": {
"owlyn": {
"command": "node",
"args": ["/absolute/path/to/owlyn-mcp/dist/index.js"],
"env": {
"OWLYN_DB_PATH": "/absolute/path/to/.owlyn/owlyn.sqlite"
}
}
}
}
Windows example:
{
"mcpServers": {
"owlyn": {
"command": "node",
"args": ["C:\\Users\\you\\Projects\\OwlynMCP\\dist\\index.js"],
"env": {
"OWLYN_DB_PATH": "C:\\Users\\you\\.owlyn\\owlyn.sqlite"
}
}
}
}
macOS/Linux example:
{
"mcpServers": {
"owlyn": {
"command": "node",
"args": ["/Users/you/projects/OwlynMCP/dist/index.js"],
"env": {
"OWLYN_DB_PATH": "/Users/you/.owlyn/owlyn.sqlite"
}
}
}
}
More host examples are in docs/HOST_SETUP.md.
Agent Instruction
Copy this into your coding agent instructions when using Owlyn:
You have access to Owlyn MCP.
At the beginning of work, call owlyn_start with the user's goal and deadline.
After meaningful progress, call owlyn_checkpoint.
Then call owlyn_should_continue.
If should_continue is true, do not stop only because the current task is complete. Pick the next highest-value low-risk task and continue.
If multiple safe next tasks exist, call owlyn_plan_next.
Before final response, call owlyn_end.
Never continue destructive, irreversible, external, expensive, deployment, credential, or approval-requiring work without explicit user approval.
See docs/AGENT_INSTRUCTIONS.md for a standalone copy-paste version.
Tools
| Tool | Purpose |
|---|---|
owlyn_start |
Start a work session with a goal, deadline, timezone, and mode. |
owlyn_status |
Return current time, elapsed time, remaining time, deadline status, and latest checkpoint state. |
owlyn_checkpoint |
Save meaningful progress, completed tasks, next tasks, blockers, changed files, validation, confidence, and risk. |
owlyn_should_continue |
Decide whether the agent should continue working. |
owlyn_plan_next |
Rank candidate next tasks by impact, risk, approval requirement, category, and fit to remaining time. |
owlyn_end |
End a session and return a final report. |
owlyn_list_sessions |
List active, completed, abandoned, or all sessions. |
owlyn_report |
Return a detailed session report without ending the session. |
Full tool documentation is in docs/TOOLS.md.
Demo Flow
User:
Work on this repo until 06:00. If you finish early, continue with safe useful improvements.
Agent starts a session:
{
"tool": "owlyn_start",
"input": {
"goal": "Improve the repo safely until morning.",
"deadline": "06:00",
"timezone": "Europe/Warsaw",
"mode": "night_shift"
}
}
Agent finishes the first task and checkpoints:
{
"tool": "owlyn_checkpoint",
"input": {
"summary": "Implemented deadline parsing and added tests.",
"completed_tasks": ["Added HH:mm deadline parsing", "Added timezone tests"],
"next_tasks": ["Add policy tests", "Document host setup", "Run release checks"],
"validation_results": ["npm test passed"],
"confidence": "high",
"risk_level": "low"
}
}
Agent asks whether to continue:
{
"tool": "owlyn_should_continue",
"input": {
"current_task_done": true,
"has_next_tasks": true,
"requires_user_approval": false,
"risk_level": "low",
"destructive_action_pending": false,
"validation_passed": true
}
}
Expected continuation signal:
{
"should_continue": true,
"reason": "The deadline has not been reached and safe next tasks are available.",
"recommended_action": "Continue with the next highest-value low-risk task. Do not stop only because the current task is complete."
}
Longer example: docs/EXAMPLE_SESSION.md.
What Owlyn Stores
Owlyn is local-first and persists session state in SQLite.
Default database path:
~/.owlyn/owlyn.sqlite
Override it with:
OWLYN_DB_PATH=/custom/path/owlyn.sqlite
Owlyn stores only what the agent sends to it:
- session goals
- notes
- checkpoints
- task lists
- file-change notes
- validation notes
- continuation decisions
Do not store secrets in goals, notes, checkpoints, task descriptions, or validation output.
Safety and Limitations
Owlyn MCP is intentionally conservative.
It should stop when work requires user approval, when destructive action is pending, when the deadline is reached, or when risk is high.
Owlyn v0.1 has:
- local-first SQLite persistence
- stdio MCP transport
- no cloud
- no telemetry
- no auth
- no browser extension
- no dashboard
- no background autonomous worker
- no shell execution by the server
- no project-file modification by the server
Owlyn MCP does not keep your agent alive by itself. It provides session state and continuation policy through MCP tools. Your host agent must call the tools and follow Owlyn's policy.
Host behavior still depends on the MCP client, model, agent instructions, context limits, tool-call limits, user approval gates, rate limits, and runtime timeouts.
Troubleshooting
better-sqlite3 native dependency issues:
Make sure Node.js and npm are installed normally for your platform. If installation fails, check that your Node version is supported and that your environment can install native npm packages.
Node version requirement:
Owlyn requires Node.js 20 or newer. Check with node --version.
OWLYN_DB_PATH usage:
Set OWLYN_DB_PATH when you want Owlyn to store its SQLite database somewhere other than ~/.owlyn/owlyn.sqlite. The parent directory is created automatically.
No active session found:
Call owlyn_start first, or pass a known session_id to tools that support it. Completed or abandoned sessions are not treated as the latest active session.
Invalid deadline format:
Use ISO datetime or HH:mm, for example 2026-07-07T06:00:00+02:00, 2026-07-07T06:00:00, or 06:00.
Roadmap
Future ideas:
- optional JSON/Markdown export
- optional session templates
- optional host-specific config helpers
- optional durable workflow integration
- optional UI dashboard
- optional per-project session profiles
- optional MCP resources for reports
- optional pause/resume
- optional session budget limits
- optional multi-agent coordination
These are not implemented in v0.1.
Development
npm install
npm run typecheck
npm run build
npm test
npm run smoke:mcp
npm run check
Release checklist: RELEASE_CHECKLIST.md.
License
MIT
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。