I built a VS Code extension that converts any technical article into a Jupyter notebook.

Problem: I kept reading blog posts and papers but forgetting everything. Copy-pasting to ChatGPT loses context.

Solution: Paste a URL, get a notebook. Then ask questions with `# %ask` — the LLM sees the full article as context.

Features: - @voyager /url <url> → generates notebook with images - # %ask --model claude-sonnet-4 <question> → AI answers with full context - Works with any Copilot model (GPT-4o, Claude, Gemini)

Built with: TypeScript, vscode.lm API, Mozilla Readability, Turndown

Marketplace: https://marketplace.visualstudio.com/items?itemName=BlackEag...

GitHub: https://github.com/blackeaglelabsai/voyagerai

Would love feedback — especially on what content types you'd want this for.

I built this over a weekend after getting tired of writing the same monitoring scripts.

You describe what you want in plain English, and it generates a working terminal dashboard with cyberpunk aesthetics using Claude + Textual.

Example: "show CPU and RAM usage" → instant TUI with live data

Key features: - Self-healing code (if it crashes, Claude fixes it automatically) - Built-in API usage tracking - Zero-install deps with uv

It's basically Lovable.dev but for your terminal. Open source, MIT licensed.

Would love feedback from the HN community!

Hey HN,

I'm a Senior SW. I built this with a friend (who is a Staff Software Engineer) as we both want to grow in the space of MCP/ Agent security and Agents Identity and Access Management (IAM).

We build MCPShark around solving security issues in MCP adoption. For example, currently, the adoption of MCP servers expose teams into security vulnerabilities such as tool poisoning, rug-pull attack etc. Thus, we built MCPShark.

MCPShark is a Wireshark like network traffic analysis tool for MCPs. It can easily integrate with developer IDEs and we intend to solve following problem.

The problem: You have ZERO visibility into what's happening. When Claude calls your MCP server, you can't see:

- What requests are being made - What data is being sent - Why tool calls fail - If there are security issues

Here's what MCPShark does: Think tcpdump + Wireshark, but specifically for MCP protocol.

Current features:

- Capture all HTTP requests/responses between IDE and MCP servers - Wireshark-like forensic analysis interface - Aggregate multiple MCP servers (HTTP and stdio) - Filter by method, status, session, server - SQLite audit logs

Status: Alpha. Still testing, rough edges. Works with Cursor, Windsurf, Claude Desktop.

We appreciate your genuine feedback. We like to solve real problems and make an impact.

- Is lack of MCP visibility a real problem for you? - What debugging/security features would help most? - Developer debugging vs security monitoring - which matters more? - Should we integrate with security tools like Invariant Labs' mcp-scan? - Above all, are we solving a real pain point. What pain points you'd like us to solve?

Links:

GitHub: https://github.com/mcp-shark/mcp-shark Desktop app (WIP): https://github.com/mcp-shark/mcp-shark-app

We're serious about agent security and IAM. Doing this part-time but committed. Open to all feedback, especially critical feedback. Happy to answer questions about implementation, architecture, or MCP in general.