An MCP server for all my rules, prompts, etc etc. Allows agents to call rules on demand.
There's no point filling 4 paragraphs of instructions for how to write tests if your current session isn't going to write a test.
Spiritually similar to Cursor's rules.
claude mcp add --transport http rules https://mcp.cianfrani.dev/mcp
{
"rules": {
"command": "npx",
"args": [
"mcp-remote",
"https://mcp.cianfrani.dev/mcp"
]
}
}Allows coding agents to lookup rules on demand within the context of their current task.
> write a test for @calc.py
⎿ Read calc.py (26 lines)
⏺ rules:get_rules_for_writing_tests (MCP)(language: "python")
⎿ - Before testing a protected method, ask "Can this logic be adequately tested through the public interface?". If yes, don't test the protected method directly.
If no, continue testing the protected method.
- Remember to write tests using the Arrange, Act, Assert pattern.
… +12 lines (ctrl+r to expand)
⏺ Write(test_calc.py)
- It's really annoying trying to carry all these little notes between projects/IDEs.
- I don't want to have to manually invoke rules.
Clearly-defined tool descriptions allow the agent to fetch rules on demand, only if they appear to be relevant.
- Patterns must be associated with a specific action. For example, "writing tests", "authoring UI", "reviewing a PR".
- The agent ultimately decides if it's going to call the tool. Sometimes it does. Sometimes it doesn't.
get_rules_for_writing_tests- Use when: writing any type of test,
- Inputs:
language(string)
get_rules_for_composing_ui
- Use when: creating new frontend components
In Claude Code, invoke prompts as slash commands.
pr-review
- Use when: you want feedback on a unit of work
- /pr-review
- See resources/pr-review.md for the full template.
- Start the server
npm run dev
- Open MCP inspector
npm run inspect