Owlyn MCP

Owlyn MCP

Time-aware work sessions for AI agents.

Category
Visit Server

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

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured