Feed aggregator

Article URL: https://teavm.org/

Comments URL: https://news.ycombinator.com/item?id=47305154

Points: 1

# Comments: 1

Every MCP server injects its full tool schemas into context on every turn — 30 tools costs ~3,600 tokens/turn whether the model uses them or not. Over 25 turns with 120 tools, that's 362,000 tokens just for schemas.

mcp2cli turns any MCP server or OpenAPI spec into a CLI at runtime. The LLM discovers tools on demand:

mcp2cli --mcp https://mcp.example.com/sse --list # ~16 tokens/tool mcp2cli --mcp https://mcp.example.com/sse create-task --help # ~120 tokens, once mcp2cli --mcp https://mcp.example.com/sse create-task --title "Fix bug" No codegen, no rebuild when the server changes. Works with any LLM — it's just a CLI the model shells out to. Also handles OpenAPI specs (JSON/YAML, local or remote) with the same interface.

Token savings are real, measured with cl100k_base: 96% for 30 tools over 15 turns, 99% for 120 tools over 25 turns.

It also ships as an installable skill for AI coding agents (Claude Code, Cursor, Codex): `npx skills add knowsuchagency/mcp2cli --skill mcp2cli`

Inspired by Kagan Yilmaz's CLI vs MCP analysis and CLIHub.

https://github.com/knowsuchagency/mcp2cli

Comments URL: https://news.ycombinator.com/item?id=47305149

Points: 1

# Comments: 0

Article URL: https://github.com/rustykuntz/termix

Comments URL: https://news.ycombinator.com/item?id=47305108

Points: 1

# Comments: 0

tmux new-session -d -s all \; set -t all mouse on \; set-option -t all destroy-unattached off >/dev/null 2>&1; for s in $(tmux list-sessions -F '#{session_name}' | grep -v '^all$'); do tmux link-window -s "$s:" -t all:; done; tmux attach -t all

Comments URL: https://news.ycombinator.com/item?id=47305101

Points: 1

# Comments: 1

Article URL: https://pmc.ncbi.nlm.nih.gov/articles/PMC7516509/

Comments URL: https://news.ycombinator.com/item?id=47305092

Points: 1

# Comments: 0

Article URL: https://github.com/ruvnet/RuView/issues/37

Comments URL: https://news.ycombinator.com/item?id=47305076

Points: 1

# Comments: 1

Article URL: https://bestofshowhn.com/

Comments URL: https://news.ycombinator.com/item?id=47305070

Points: 1

# Comments: 0

Article URL: https://seedancev2.ai

Comments URL: https://news.ycombinator.com/item?id=47305068

Points: 1

# Comments: 0

Article URL: https://www.notebookcheck.net/Imprisoned-hacker-hints-GTA-6-source-code-could-leak-threatening-release-date-delay.1245446.0.html

Comments URL: https://news.ycombinator.com/item?id=47305064

Points: 1

# Comments: 0

Article URL: https://www.bbc.com/news/articles/cwy92qpv424o

Comments URL: https://news.ycombinator.com/item?id=47305061

Points: 1

# Comments: 0

Entropick is a vLLM plugin (also works with Transformers and llama.cpp) that swaps out the PRNG in token sampling with true randomness from external sources. I am using it with the Crypta Labs QCicada QRNG device: https://cryptalabs.com/quantum-random-number-generator/

The core question: does the source of randomness in token sampling matter? Standard inference uses deterministic PRNGs. Entropick lets you swap in physical entropy and run controlled experiments across different sampling configurations to find out.

The pipeline has 13 composable stages — temperature, top-k, top-p, min-p, Mirostat, Gumbel-Max, DRY repetition penalty, logit perturbation, adaptive injection, and more. Each can be toggled and tuned independently via env vars or per-request. The goal is to make it easy to experiment with every permutation of token selection algorithms and entropy sources, then compare results with real statistics.

When the model produces logits, entropick fetches raw bytes from your entropy source just-in-time (the measurement happens after logits exist, never before), amplifies thousands of bytes into a uniform float via z-score statistics, and selects a token from a CDF. The signal amplification is designed so that even a tiny statistical bias in the raw bytes produces a measurable shift in token selection.

To be clear — this project makes no claims about whether quantum randomness impacts LLM output. It's infrastructure for testing that question rigorously. If the answer is no, you still get a well-engineered entropy plugin with a flexible sampling pipeline.

GitHub: https://github.com/ereid7/entropick

Comments URL: https://news.ycombinator.com/item?id=47305051

Points: 1

# Comments: 0

Rainy Updates started as a deterministic dependency review tool for Node monorepos and CI.

With v0.7.0, it expands into cross-stack supply-chain review and attestation policy checks for local workflows, CI gates, and MCP-compatible agents.

This release adds: - cross-stack scanning for Docker, GitHub Actions, Terraform, and Helm - normalized findings with riskLevel, policyAction, and recommendedAction - attestation verification with deterministic verdicts: allow, review, or block - non-mutating MCP tools for supply-chain and attestation workflows

The goal is to make software change review more deterministic across dependencies, supply-chain exposure, and release trust posture.

Would love feedback on: - whether this feels meaningfully different from PR-first dependency automation - what’s missing for real CI usage - whether the local/MCP review model is actually useful

Comments URL: https://news.ycombinator.com/item?id=47305032

Points: 1

# Comments: 1

I built SocialProof after noticing that most freelancers and small agencies have a bunch of happy clients but almost no written testimonials on their site – because asking for them is awkward and the follow-through rate is low.

SocialProof gives you a single shareable link. You send it to a client, they fill a short form (name, text, optional photo), you approve it in your dashboard, and it appears on any page you embed the widget on.

No login required for clients submitting testimonials. The collection form is open – you can try submitting one here: https://socialproof.dev/c/frm_iOEPvPH9TzPypGMFNhYPV

Stack: Cloudflare Workers + D1 + Pages, React frontend. Free to start, paid plans for multiple widgets.

Happy to answer questions about the technical decisions or the problem space.

Comments URL: https://news.ycombinator.com/item?id=47305025

Points: 1

# Comments: 0

Article URL: https://lord.io/spreadsheets/

Comments URL: https://news.ycombinator.com/item?id=47305011

Points: 1

# Comments: 0

Article URL: https://everytool.solutions/

Comments URL: https://news.ycombinator.com/item?id=47305002

Points: 1

# Comments: 1

Article URL: https://windyroad.com.au/blog/making-work-in-progress-visible-to-your-ai-agent

Comments URL: https://news.ycombinator.com/item?id=47305001

Points: 1

# Comments: 0

Build slacrawl based on discrawl to use local storage data from Slack to wiretap messages to feed into my openclaw instance realtime.

Comments URL: https://news.ycombinator.com/item?id=47304989

Points: 1

# Comments: 0

Article URL: https://www.nytimes.com/2026/03/08/nyregion/nypd-shield-history.html

Comments URL: https://news.ycombinator.com/item?id=47304986

Points: 1

# Comments: 0

Article URL: https://notbor.ing/words/the-joy-of-building-slow

Comments URL: https://news.ycombinator.com/item?id=47304969

Points: 2

# Comments: 1