{"id":6449,"date":"2026-08-06T09:00:00","date_gmt":"2026-08-06T07:00:00","guid":{"rendered":"https:\/\/storiesonboard.com\/blog\/mcp-skills-claude-codex"},"modified":"2026-08-06T09:00:00","modified_gmt":"2026-08-06T07:00:00","slug":"mcp-skills-claude-codex","status":"publish","type":"post","link":"https:\/\/storiesonboard.com\/blog\/mcp-skills-claude-codex","title":{"rendered":"MCP Skills in Claude and Codex: A Practical Comparison"},"content":{"rendered":"

MCP skills<\/strong> are often described as if they were one feature. They are really two complementary layers. The Model Context Protocol (MCP) gives an AI agent standardized access to tools, data, and external systems. A skill gives the agent a reusable way to perform a task: the instructions, decision points, checks, and handoffs that turn access into a repeatable workflow.<\/p>\n

Claude and Codex support both ideas, but they package, discover, and invoke skills differently. This guide compares the two approaches using a real five-skill StoriesOnBoard workflow and a deliberately small product brief called Daily Reset.<\/p>\n

MCP and agent skills solve different problems<\/h2>\n

MCP is an open standard<\/a> for connecting AI applications to external systems. An MCP server can expose tools the model may call, resources the application can place in context, and reusable prompts the user can invoke. In product work, that can mean reading a story map, discovering personas and releases, creating cards, or updating acceptance criteria in the system where the team actually works.<\/p>\n

A skill is the operating procedure that tells the agent how to use those capabilities well. It can say: challenge vague assumptions before creating a backlog; show a plain-text preview before any write; create goals before steps and stories; continue after a partial failure; then read the board back and verify the result.<\/p>\n\n\n\n\n\n\n
Layer<\/th>\nQuestion it answers<\/th>\nTypical contents<\/th>\n<\/tr>\n<\/thead>\n
MCP<\/strong><\/td>\nWhat systems can the agent access or change?<\/td>\nTools, resources, prompts, authentication, transport<\/td>\n<\/tr>\n
Skill<\/strong><\/td>\nHow should the agent complete this job?<\/td>\nInstructions, checks, templates, references, scripts, approval gates<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

This is why \u201cMCP skill\u201d is useful shorthand but an imprecise technical term. A skill can work without MCP and produce a text deliverable. An MCP connection can also work without a custom skill, leaving the model to improvise the process. The strongest setup combines the two: the skill supplies the method, while MCP supplies live context and actions.<\/p>\n

\n \"Diagram
A skill defines the workflow; MCP connects that workflow to real tools and data.<\/figcaption><\/figure>\n

How skills work in Claude<\/h2>\n

In Claude Code, a skill is a directory with a SKILL.md<\/code> entry point and optional supporting files. Claude can load a relevant skill automatically from its description, or the user can invoke it directly with \/skill-name<\/code>. According to the Claude Code skills documentation<\/a>, the body loads only when the skill is used, so detailed procedures do not consume the working context on every turn.<\/p>\n

Claude supports several skill locations:<\/p>\n