Scout is the 10x Developer OS for the AI era. Use the v1 Cloud Dashboard to manage your GitHub contributions, and the v2 Local MCP to safely orchestrate AI agents directly in your IDE.
The Scout v1 Cloud Dashboard is publicly hosted and ready to use:
π https://rahul-pamula.github.io/Open_Source_Scout/
Just bring your own Supabase backend and GitHub token. Setup takes about 5 minutes.
π New in v2.2.0: Scout MCP Server 128-PR Enterprise Overhaul! Complete with Git worktree concurrency, universal multi-language test validation, TF-IDF dynamic skills, real-time MCP resources, and cross-platform CLI diagnostics. Read the v2.2.0 Release Notes, MCP Architectural Specification, and Client Setup Guide (Cursor, Claude, Cline, Windsurf).
Scout is built for modern developers who use AI. It consists of two completely independent engines:
A Bring-Your-Own-Backend (BYOB) platform that connects to your personal Supabase instance.
- Issue Scanner: Discovers open-source issues on GitHub that match your developer profile.
- AI Evaluator: Evaluates each issue with AI to tell you the difficulty and skill match.
- Mission Control Pipeline: Tracks your contributions through a visual Kanban pipeline (Discovered -> Claimed -> Assigned -> Review -> Merged).
- Interview Prep Memory: Directly attach and manage your personal Google Docs technical notes for every merged contribution, ensuring zero human context decay before technical interviews.
A production-grade, 100% offline local execution harness that connects your AI assistant (Cursor, Claude Desktop, Cline, Windsurf) via the Model Context Protocol (MCP):
- Git Worktree Isolation: Isolates all AI code execution inside hidden
git worktreeson dedicated branches so your main branch is never corrupted. - Sandboxed Execution Harness: Subprocess tracking via
tree-kill, configurable timeouts, path traversal defense, and sensitive token redaction. - Universal Multi-Language Validation: Auto-detects and runs test suites across Node (npm, pnpm, yarn, bun), Deno, Python, Rust, Go, Java/Kotlin, .NET, and Make.
- Dynamic TF-IDF Skills Discovery: Auto-discovers and injects workspace context and style rules into agent prompts based on task descriptions.
- Real-Time MCP Resources & Telemetry: Exposes live diffs, status updates, timeline events, progress notifications, and hardware telemetry profiling (
peak_rss_bytes). - Comprehensive CLI Tooling: Includes
scout-mcp doctordiagnostics and instant integration with Cursor, Claude, Cline, and Windsurf.
GitHub hosts tens of millions of issues, enforces strict Search API limits (30 requests/min), and contains significant noise.
"Instead of naively crawling millions of GitHub issues, our system uses a funnel architecture with Push-Down Filtering. We push query constraints directly to GitHub's search index to cut millions down to hundreds, pass those through a zero-latency deterministic edge filter, deduplicate against our local pipeline state, and only run expensive AI evaluations on claimed candidates."
flowchart TD
A["GitHub Universe (100M+ Issues)"] -->|1. Push-Down Query Filtering| B["~50 Fresh Candidates"]
B -->|2. Deterministic Edge Filter| C["Eligible Unassigned Issues"]
C -->|3. Local Pipeline Deduplication| D["10 High-Signal Scouted Issues"]
D -->|4. Lazy On-Demand Evaluation| E["Autonomous AI Execution (Scout MCP)"]
- Push-Down Filtering (Upstream Index Querying)
Rather than downloading huge feeds, Scout constructs scoped search queries (e.g.,
is:open is:issue label:"good first issue" language:typescript sort:updated-desc) directly against GitHub's search engine. GitHub's inverted indexes do the heavy lifting, instantly narrowing down 100M+ issues to a compact pool of ~50 candidates. - Deterministic Edge Filtering (Zero-Latency Rule Engine)
The candidate batch passes through our Deno Edge Function's
FilterEngine, instantly discarding low-signal issues without calling an LLM:- Rejects issues that already have assignees (
isAssigned). - Drops blacklisted tags (
wontfix,duplicate,invalid,spam,stale,question). - Rejects low-effort submissions (bodies
< 20characters).
- Rejects issues that already have assignees (
- State Deduplication (O(1) Pipeline Filtering)
Candidates are cross-referenced with your Supabase pipeline via an
O(1)Set lookup. Issues already claimed, assigned, merged, or dropped are filtered out so you only see fresh opportunities. - Lazy AI Evaluation (Token Cost Optimization) Expensive LLM reasoning and MCP agent orchestration are never run across raw discovery feedsβthey are triggered lazily, only when an issue is claimed for autonomous execution.
Since Scout v1 is a BYOB architecture, you host the backend yourself on Supabase (free tier is perfect).
# Run the interactive setup CLI
npx open-source-scout setupThe CLI will automatically apply PostgreSQL migrations and deploy the Deno Edge Functions to your Supabase project. Then, just visit the hosted dashboard and log in with your keys.
Add this JSON snippet to your AI assistant's MCP configuration file (Cursor, Cline, Claude Desktop):
{
"mcpServers": {
"scout": {
"command": "npx",
"args": ["-y", "open-source-scout-mcp"]
}
}
}For a deep dive into the architecture, security guardrails, and deployment guides, check out the Official Documentation.
Contributions are welcome! Please read our Contributing Guide and Code of Conduct.
This project is licensed under the MIT License - see the LICENSE file for details.
Open Source Scout includes an integrated Workspaces OS designed for developers and interns to organize external tools, documentation, repositories, and credentials.
- 1-Click App Launcher: High-density mobile-style squircle launcher.
- Zero-Knowledge Vault: Client-side AES-256-GCM encryption for passwords.
- Daily Standup Generator: Format tasks and links directly for Slack and Teams.
graph LR
A[Uncontested Issues] --> B[Scout AI Discovery]
B --> C[Worktree Sandbox]
C --> D[Multi-Language Test Validation]
D --> E[Merged PR & Standup Report]
