Tag

Mcp

All articles tagged with #mcp

technology29 days ago•74 min saved

MCP overhaul strips sessions, shifting state to per-call metadata for lean servers

The Model Context Protocol (MCP) gets its biggest update since launch: persistent sessions and the initialization handshake are removed, with identity and capabilities now carried in per-call metadata and a new server/discover method added. This enables MCP servers to operate more like stateless HTTP services behind gateways, while essential state is handled via explicit handles (e.g., carts, task IDs). The change aims to reduce infrastructure complexity under a pay-as-you-go model, with deprecation guidance and migration paths to help developers adapt.

via The New Stack|

#api-management #mcp #migration

technology3 months ago•70 min saved

OpenClaw at 300k stars as Spark arrives: the hosted vs. self-hosted AI agent showdown

OpenClaw, a self-hosted AI agent that users could run on their own hardware, surpassed 300,000 GitHub stars, while Google unveiled Spark, a 24/7 hosted Gemini-based agent that runs on Google Cloud and threads into Gmail, Docs, and Sheets. Both are converging on MCP for tool connectivity, but differ in substrate: OpenClaw emphasizes control and ownership of credentials and workflows, whereas Spark offers hands-free convenience with Google hosting the runtime and context. The choice boils down to privacy and autonomy versus ease of use and seamless integration, signaling a two‑tier future for AI agents.

via The New Stack|

#ai-agents #gemini-35-flash #mcp

ai7 months ago•66 min saved

Anthropic opens MCP to interactive UI apps inside Claude chat

Anthropic released a major open extension to the MCP protocol that lets MCP servers render interactive, app-like UIs within chat interfaces. The MCP Apps framework enables third‑party tools and services to embed widgets and workflows directly in Claude, with early partners including Amplitude, Asana, Box, Canva, Figma, Slack and Salesforce planning integration. The feature builds on MCP‑UI and the ChatGPT Apps SDK, enabling scenarios like inline file previews, draft editing in Slack, dashboards, configuration wizards, and real‑time monitoring. Security guards include sandboxed iframes, host review of HTML content, auditable UI‑to‑host messages, and explicit user approval for UI‑initiated tool calls. The standard is open, so other chat providers can adopt it and build apps atop MCP.

via The New Stack|

#ai #anthropic #claude

technology1 year ago•6 min saved

MCP Aims to Reduce Developer Distractions and Boost Focus

Recent research shows developers spend only 16% of their time coding, with the rest on operational tasks, largely due to frequent context switching between tools. The Model Context Protocol (MCP), introduced by Anthropic, aims to streamline workflows by integrating AI assistants directly with external tools and data sources, potentially transforming software development into a more efficient, all-in-one process. However, MCP faces security, scalability, and practical adoption challenges that need addressing before widespread enterprise use.

via VentureBeat|

#ai-coding-assistants #context-switching #developer-productivity

technology1 year ago•39 min saved

GitHub Copilot Enhances Coding with Agent Mode and Web Browser

The article discusses the complexities and practical benefits of using GitHub Copilot's Agent Mode and MCP in software development, emphasizing that these tools are primarily aids for decision-making and automation rather than replacements for comprehensive engineering processes. It highlights real-world scenarios where AI accelerates coding tasks, but also cautions about over-reliance and organizational issues, advocating for thoughtful integration and communication.

via Hacker News|

#agent-mode #github-copilot #llms