Tech News
← Home  ·  All topics

Mcp

63 GoKawiil briefs on this topic

Open-source author proposes shared memory spec for AI coding agents

A developer behind the open-source tool LoreKit argues that AI coding assistants currently store 'memory' in incompatible formats, including AGENTS.md files, Cursor's .mdc rules, Anthropic's memory tool directories, GitHub Copilot's repo facts, and various MCP-based knowledge graphs. The piece surveys these formats and existing proposals, drawing a parallel to observability tooling before OpenTelemetry unified it with a common data model.

Researchers find ChatGPT can silently route confirmation codes between two connectors

A security test showed that when a ChatGPT account had two connectors pointing to the same MCP server, the model confirmed an action's one-time code via one connector but redeemed it through the other, without informing the user. The server rejected the mismatched call because the code was cryptographically bound to the first connector's OAuth client, but the switch itself happened silently in every logged attempt.

New open-source tool 'Agentic Identity Broker' manages OAuth delegation for AI agents

A developer has released Agentic Identity Broker, an open-source project shared on Hacker News that lets AI agents get scoped, revocable access to services like Google, GitHub, Databricks and Linear without being handed user credentials. It sits in the request path via gateways such as Agentgateway, handling OAuth2 delegation chains, token vaulting, consent management and audit logging on behalf of agents and MCP servers.

Finic launches Kairos, an open-source fraud and risk decisioning platform

Finic has released Kairos, an open-source real-time fraud and risk decisioning system that evaluates sign-ups, logins, orders, payments and payouts, returning an approve, step up, review or block decision along with a risk score and supporting signals. The platform includes an MCP server so AI coding agents like Claude Code, Cursor or Codex can build workflows, propose model variants and tune detection models, alongside traditional machine learning options like logistic regression and gradient-boosted trees, with SDKs for TypeScript and Python plus self-hosting via Docker.

Octomind launches as open-source CLI runtime for autonomous coding agents

Octomind is a new open-source, CLI-first AI coding agent runtime that gives language models direct access to a repository to read and write files, run shell commands, search code, and delegate to sub-agents via MCP tools. It can run interactively, through piped input, as a background daemon, over WebSocket, or as a sub-agent within another agent framework, and supports policy rules written in TOML plus custom scripts for guardrails.

Nvidia releases open-source Auto Ontology library for AI agents

Nvidia has launched Auto Ontology, an open-source library under the Apache License 2.0 that builds a context layer from enterprise metadata to help AI agents interpret data accurately. The tool reconciles differing data definitions across systems, generates knowledge graphs, and translates natural-language questions into SQL queries using NVIDIA NIM. It exposes this context to agent frameworks via MCP, CLI, and API, and is not currently accepting external contributions.

BBRouter launches 3D breadboard simulator for Ben Eater's 8-bit computer

BBRouter is a new tool that recreates Ben Eater's well-known 8-bit computer build as an interactive 3D breadboard simulation. It also ships an MCP server that lets AI assistants such as Claude connect, open, edit and run a user's breadboard projects directly, after a one-time sign-in and authentication step.

Sketch Stadium MCP tool lets AI coding agents play Pictionary live

A new MCP server called Sketch Stadium lets AI agents such as Claude Code, Codex and Claude Cowork join live Pictionary-style games. Users connect their agent via an MCP URL, authorize access in a browser, and then watch a live room where the agent draws or guesses words in real time, with humans able to type guesses alongside it.

DoorDash: AI agents placed only 'several hundred' orders so far

DoorDash CEO Tony Xu said on the company's Q3 earnings call that third-party AI shopping agents like Muse, Instinct and Fo have generated only a few hundred total orders, compared to the roughly 3 million deliveries DoorDash completes daily. Xu said most users still order through DoorDash's own app, while the company continues building its own agent tools, including a text-to-order agent, a CLI agent, and the Ask DoorDash chatbot.

phpClaw launches as open-source PHP runtime for AI agents

A new open-source library called phpClaw was released on Show HN, offering PHP developers a ready-made runtime for building AI agents. It includes tool integration, memory, security guards, and a ReAct reasoning loop, and works with multiple LLM providers including Anthropic, OpenAI, Groq, Gemini, Mistral, DeepSeek, and Ollama without requiring any framework dependencies.

New Open-Source Tool PaperPilot Offers Local PDF Editing for AI Agents

PaperPilot is a Rust-based PDF toolkit that runs locally as a CLI, browser WASM app, desktop app, REST API, or MCP server for AI assistants like Claude Desktop and Cursor. Version 0.6 supports merging, splitting, rotating, compressing, OCR, watermarking, signature validation, PII scanning, and Excel conversion, all without uploading files to the cloud or relying on Chromium or C++ runtimes.

ADRs4AI launches Git-based toolkit for recording human-AI decision reasoning

A new open-source project called ADRs4AI offers a template system, VS Code extension, and iOS app for capturing architecture decision records (ADRs) generated during human-AI collaboration sessions. The toolkit includes seed templates, an inbox protocol, task-runner recipes, and an MCP server so AI agents can query the stored decision history, with records versioned directly in Git.