agent-accesslisted
Install: claude install-skill ShlomiPorush/agent-skills
# Agent access
This skill adds agent connection to a web product. An AI agent acts in the product for one person, after that person approves it in the browser, and only within what the person granted. It distils two products that built the pattern and hardened it over many review rounds: n8n-ops (TypeScript, Hono, `oidc-provider`) and Comply (Python, FastAPI, OAuth written by hand).
## First: check that the request is this one
Use this skill only when the user asks for one of these in a product they build:
- AI agents connect to the product and act for a person who approved them;
- an MCP server whose clients get access through OAuth that a person approves;
- an agent skill, served by the product, whose command connects with a code a person approves;
- the n8n-ops or Comply agent-access pattern in another project;
- bringing a product that already has agent connection up to this standard (see "An existing implementation").
One channel alone is fine: build the MCP server, the agent skill, or both, as the request needs. When the product already runs an OAuth authorization server, build on it.
These are not this skill: adding sign-in or social login, a device-code sign-in for people using the product's own CLI, building an OAuth client, an MCP server for a local tool, a skill that calls an API with an API key, using the `n8n-ops` or `comply` skills to work in those products. If the request is one of those, do not use this skill.
## What you build
Two channels share one p