Skip to content

Getting Started with mcp-guide

So you've got mcp-guide installed and running. Now what?

This section answers the most important question: how do I actually use it?

Guide MCP Purpose

At its core, mcp-guide serves instructions to your AI agent. Think of it as a structured way to tell your agent how to work on your project - providing information on coding standards, workflow steps, project-specific context - all delivered on demand.

While it can do more than that, serving agent instructions is the fundamental purpose. Everything else builds on this.

The guide:// Interface

Your primary interface is the guide:// URI scheme. Some clients also expose equivalent prompt syntax, but the URI form is the stable, portable option.

When you use a guide:// reference with an expression that contains just a category name, you get back all the documents in that category that match the patterns set for that category. The agent receives these documents and acts on them according to their type.

Basic Usage

Common Examples:

Reference What It Does
guide://commit Provides commit message instructions based on changes in your working tree
guide://pr Provides GitHub pull request description instructions for changes compared with your main branch
guide://review Instructions for the agent to self-review code changes
guide://guidelines Composite request presenting coding standards, language-specific guides, and project context

Guide URIs

For clients that support prompts, guide://commit may also be available as a prompt-style form. Commands like guide://_status remain the canonical documented syntax. If the agent can't read URIs directly, the read_resource tool works as a fallback.

See Guide URIs for the full details.

Project Detection

mcp-guide needs to know which project you're working in. Most CLI agents handle this automatically, but some (like Codex) require the agent to call set_project with the project path first. See Installation — Project Detection for details.

Document Categories and Collections

Categories

A category is a named group of documents with an associated directory (usually the same as the name), an optional description and zero or more patterns. The category name is limited to 30 printable unicode characters and cannot start with an underscore or hyphen and cannot contain whitespace.

When you reference a category without specifying a pattern, you retrieve documents matching the patterns.

Patterns are "globs", and may contain '', '?', '*' and path separators. However, they are usually used to match the basename of documents (filenames without any extension). The document matcher automatically tries several extensions by default, favouring markdown, mustache templates and text files.

When referencing a category, you can also override the default patterns:

guide://lang/python

Retrieves the python language documents from the lang category.

guide://docs/workflow+tracking

Retrieves documents matching BOTH workflow AND tracking patterns in the docs category.

What you're providing is an expression, which takes the form of a comma-separated list of categories, with optional patterns for each, separated by '+'. When retrieved, all matching documents are returned to the agent.

Collections

Collections are groups of category expressions - convenient shortcuts or "macros" for commonly used combinations.

Instead of typing guide://guide,lang/python,context/project,checks every time, you might have a collection called guidelines that includes all of those.

Collections make it easy to provide targeted context for specific tasks without remembering complex expressions.

See Categories and Collections for managing these.

Document Types

What the agent does with documents depends on their type, set in the document metadata:

  • agent/instruction - Directives the agent must follow (most common)
  • agent/information - Context and reference material for the agent
  • user/information - Content displayed to users

When invoking multiple patterns, be aware that mixing document types may not give the desired result. Most documents served via guide:// are agent instructions - this is the core of what mcp-guide does.

Default Profile

When you start mcp-guide in a new project, it creates a default profile with the following categories:

  • guide - General Guidelines and Instructions
  • docs - General Uncategorised Documents
  • lang - Language-specific guides (initially empty)
  • context - Project context (initially empty)
  • checks - Testing and validation instructions (initially empty)
  • review - Code review instructions

Before these can become useful, you will need to add patterns: - Add your programming language to lang - Add project context (tracking method, workflow approach, issue tracking etc.) - Add testing frameworks and special instructions to checks - Configure review depth and output format in review

Many provided documents are opinionated about development approach and direction. Use what works for you, ignore the rest.

To discover available documents in a category you can simple ask the agent to list all files available in that category.

Special: INSTRUCTIONS.md

The checks category includes a brief document instructions that tells the agent to look for INSTRUCTIONS.md in your project root. This file contains edicts the agent MUST obey during development - rules specific to your project or agent that curb common misbehaviors.

The checks phase in workflow mode is a good place to remind the agent of these instructions before formal examination of changes. Also consider appending checks/instructions to your default guidelines collection.

Adding Profiles

Other profiles can be added on demand - they are additive to your existing configuration. See Profiles for details.

Commands

Commands are specialized documents that tell the agent to do something specific or present useful information using a templated layout. They are special in that they use a _ prefix in the guide:// scheme: guide://_<command>.

Many commands accept these common arguments:

Argument Description
-h, --help Show help
-v, --verbose Detailed output
-q, --quiet Minimal output
-f, --force Skip confirmations
-d, --debug Debug information
-t, --table Tabular format

Not all commands support all arguments. Use guide://_help/<command> to see what's available for a specific command.

Useful Commands

Command Description
guide://_project Information about the current project, categories, and collections
guide://_status Current project name, active tasks, and workflow state (if enabled)
guide://_agent Information about the running agent and its attributes
guide://_create/category Create new categories
guide://_create/collection Create new collections
guide://_help List available commands
guide://_help/<command> Detailed help for a specific command

Note: guide://_project, guide://_status, and guide://_agent may show incomplete information if the agent hasn't yet requested or determined certain details through tool use. Running them a couple of times or using a general guide://<category> request will usually shake this out immediately.

First Run - Self Installation

When mcp-guide first runs, it creates:

  • Configuration directory:
  • macOS/Linux: ~/.config/mcp-guide/

  • Windows: %APPDATA%\mcp-guide\

  • Configuration file: config.yaml (shared by all projects)

  • Document root (docroot): docs/ directory within the configuration directory

  • Standard templates: Copies of all default documents and templates

If you want to add custom documents, place them in the docs/ directory. Each category has a designated subdirectory (usually matching the category name). Files within these directories become available to serve via the MCP.

The docroot is on the MCP server's filesystem - if running in Docker or on a remote server, that's where the files live, and are not available on your local machine unless mapped, or "mounted".

Next Steps

Now that you understand the basics:

The key thing to remember: mcp-guide serves instructions to your AI agent. Everything else is about organising and delivering those instructions effectively.