与「mcp」相关的搜索结果

Alma 0.4.672 支持了 Poppy（Personal Agent Protocol）： · 自动发现公司的 /.well-known/poppy.json · 访问它的 API / 页面 / MCP server 时带上签名的 Poppy-Session 头，商家知道这是替用户办事的个人 agent，而不是匿名爬虫 · UCP 结账：带上 Alma 的 UCP 平台 profile，实测 Allbirds 的 Shopify UCP 端点协商出 checkout / cart / order 等全部能力 让 agent 正大光明地去购物 🛍️

靠，手机上想用 AI，ChatGPT 一个 App、Gemini 一个 App、DeepSeek 又一个，自己买的 API Key 还没地方填，桌面全是聊天框。 这个开源的安卓客户端把 OpenAI、Anthropic、Gemini 兼容的接口全收进一个 App，自己填 Key 随便换模型，MCP、Skills、联网搜索都能接，还塞了个 Linux 工作区让 Agent 在手机里跑命令，甚至能让 AI 直接替你点手机。 GitHub 快 9000 星，安卓党这回不用在一堆 AI App 之间来回切了。

GitHub 狂揽 10.7 万星的「本地版 Claude Design」完整实战指南来了 内置 151 套大厂品牌系统与 277 个插件，连上 Claude Code 就能一口气产出原型、PPT 与宣传片 6 步实操流程、防返工提示词与 MCP 跨仓交接命令请看下文👇

Now that Solana has abandoned the expedition into geographically distributed MCP, it's high time we revisited Solana's broader approach to geographic distribution. Geographically distributed MCP was always weak at best. Even if you ignored the insane latency involved in going from transaction submission to assembled blocks and execution results, it was not a good CR mechanism. MCP's CR guarantees depend heavily on users' ability to send transactions to at least two proposers, and this imposes a trade-off between slot duration and geographic decentralization. If the ratio is not managed, there will be instances where your transaction cannot make it to more than one leader within the slot time. Sending it to a leader you trust defeats the purpose of MCP; it has the same bound on geographic decentralization and it's unclear whether all the extra complexity is justified by this weaker guarantee compared to the status quo. Constellation's leader colocation fixes that completely. Users can now send to all the leaders with ~the same latency, and the Nash equilibrium is to include the top fee-paying transactions. However, I think the current Constellation is not aggressive enough. Solana is rechoosing its spot on the physical distribution spectrum, and current Constellation leaves some chips on the table. In the coming days, I'll write up a design I've been working on that I think maximizes what's possible. I'm talking fully onchain orderbooks that are competitive with centralized systems. All said, the most important takeaway from Roger's talk and the entirety of today is that Solana's people and culture are its greatest assets. There is a pragmatism and relentless dedication to winning that I haven't seen anywhere else!

4 open-source GitHub repos for design workflows: ✧ OpenPencil: local .fig editor + MCP server ✧ Penpot: native CSS Grid & SVG design platform ✧ Graphite: Rust node-based vector & motion engine ✧ Style Dictionary: multi-platform token compiler

Who decides the answer to a prediction market? Usually the creator. Or an AI on a server nobody can check. So we built Noa Court - an AI resolver for Noa, live on NEAR mainnet. In the video we ask Gemini a vague question: "will the NEAR hackathon get more than 2000 projects?" - It finds the real source (Nearhackathon API) and samples it live: 1,803 right now - Turns the vague question into precise, hashed rules - We review the market and sign with a NEAR wallet - Market #27# goes live on Noa, published with readable rules How it resolves: - One Court contract, one pinned resolver running in an @out_layer TEE - Anyone can trigger it. Nobody can pick the answer - Numbers? Decided by code - Words? 3 NEAR AI judges must all agree and quote the evidence Asked "did a crypto hack make the news?", all 3 judges quoted a headline that never says "hack". It refuses to guess: - Prompt injection - flagged - Dead or changed source - void + refund Works with prices, NEAR tokens (powered by Intear & JustHoot), news feeds, web pages, any JSON API. Plug it into any market, contract or agent. Add the MCP to Claude, ChatGPT, Gemini - any AI that speaks MCP: Code: What would you never let an AI resolve? Built at NEAR HACKATHON @NEARProtocol @NearBuilders #NEARHackathon#

卧槽，装个插件等于请了个贼进门😅 11 月 13 日，npm 上一个叫 karma365-mcp 的包被 GitHub 标为恶意软件。名字蹭着 MCP 和 Claude 的标签，专钓装插件的人。 装过的机器，按被入侵处理：先清掉，再换密钥😬

PageIndex 0.6.0 把文档索引、问答和 Agent 工具都收进了一个客户端，本地和云端走同一套接口。 本地默认用 Flash 引擎，照着 PDF 的版式建树形索引，不用向量数据库，也不用 PageIndex API key。模型调用还是走你自己配的后端。 OpenAI / Anthropic 的工具和 Claude MCP 接口也都能直接导出。在做 PDF 问答或文档 Agent 的，可以拿这套接口试试，看接入能不能省点事。

NVIDIA released Auto Ontology, an open-source library that builds a semantic layer from your enterprise metadata and exposes it to agents via MCP. Vendor tests claim **90% faster ontology building**. The problem is familiar: agents that write SQL against your data use their own definitions, not the enterprise's. A "revenue" the agent calculates doesn't match the one in the financial report. Auto Ontology builds a semantic layer from existing metadata (schemas, tables, columns, relations), reconciles definitions across systems, and exposes it to agents via MCP, CLI, or API. Nine MCP tools are provided to query the catalog, validate a query, or retrieve certified definitions. What the account takes from this: the 90% gain comes from NVIDIA's own tests, not an independent measurement. The library automates construction, not governance. Business definitions, access rights, and auditing remain your responsibility. The gesture: clone the repository, point it at a test schema, and see what entities and relations it extracts. The README shows the complete flow in ten minutes.

Your AI agent can now build, light and render a 3D scene end to end, and the engine it drives is free and open source. 🔥 It's called 67js, by CodePawl. You prompt Claude Code, Codex or Cursor, and the agent works through an MCP server: places objects, sets lights and fog, checks PNG previews, then renders a still or an MP4. It's catching on fast: 633 GitHub stars and about 4,100 npm downloads last week. Setup is one prompt to your agent, or a Claude Code plugin. MIT licensed, needs Node 20.19+ and a local Chrome, and falls back to CPU if you don't have a GPU. Good call on safety: the MCP server accepts no arbitrary code and no remote asset URLs, only scene data and the GLB files you point it at. The catch is it's pre-1.0, ships a new minor version about every week, and a minor can break the API. Great for product shots and explainer clips today, too early to build a pipeline on. 🔗 Link in the first reply 👇