MCP & Agent Fixes

Browse all 2 primary-source-verified engineering guides in MCP & Agent Fixes. New articles added to this category appear here automatically in reverse chronological order.

2 Verified GuidesSorted by Latest Verification Date

Illustrative meanings of Ollama processor placement and documented automatic context tiers; not a measured terminal session. Mobile reference view.

Primary Source: Ollama Docs

MCP & Agent Fixes11-09 Verified

Ollama GPU Offload & Context Length Tuning: Fixing CPU Fallback, OOM & Slow Token Generation

How to diagnose partial CPU/GPU memory splits with ollama ps, override Ollama's VRAM-based default context lengths (4k/32k/256k), and configure OLLAMA_CONTEXT_LENGTH, num_ctx, and CUDA_VISIBLE_DEVICES.

Editable Windows MCP launch reference using cmd /c, real runtime lookup and clearly labelled path placeholders. Mobile reference view.

Primary Source: MCP · Node.js Docs

MCP & Agent Fixes11-09 Verified

Fixing Claude Desktop MCP "spawn npx ENOENT" & Server Disconnected Errors on Windows and macOS

Why local Model Context Protocol (MCP) servers fail to launch with spawn npx ENOENT or Server disconnected in Claude Desktop, and how to fix claude_desktop_config.json on Windows (cmd /c wrapper), macOS, and Claude Code.