git-mcp exposes four read-only MCP resources that map common Git views to URI-addressable endpoints. Resource-aware clients can subscribe to these URIs and receive structured JSON rather than calling tools.

Available Resources

URI	Description
git+repo://status/{repo_path}	Working tree status (staged, unstaged, untracked)
git+repo://log/{repo_path}	The 20 most recent commits
git+repo://branches/{repo_path}	All local branches with tracking info
git+repo://diff/{repo_path}	Unstaged and staged changes

URI Format

The {repo_path} segment is the absolute path to the repository, URI-encoded. For example:

git+repo://status//home/user/myproject
git+repo://log//Users/dan/dev/app

On Windows, encode the drive letter path: git+repo://status/C%3A%5CUsers%5Cdan%5Cdev%5Capp

Resource Contents

git+repo://status/{repo_path}

Returns a JSON object mirroring the output of git status --porcelain=v2. Fields include:

• staged — files in the index (key: file path, value: status code)
• unstaged — files modified in the working tree
• untracked — new files not yet tracked
• branch — current branch name or HEAD if detached
• tracking — upstream ref and ahead/behind counts

git+repo://log/{repo_path}

Returns a JSON array of up to 20 commit objects, each with:

• hash — full commit SHA
• short — abbreviated SHA (7 chars)
• author — { name, email }
• date — ISO 8601 timestamp
• message — commit subject line
• body — commit body (may be empty)

git+repo://branches/{repo_path}

Returns a JSON array of branch objects, each with:

• name — local branch name
• current — true if currently checked out
• upstream — tracking remote ref (or null)
• ahead — commits ahead of upstream
• behind — commits behind upstream

git+repo://diff/{repo_path}

Returns a JSON object with:

• staged — unified diff of staged changes
• unstaged — unified diff of unstaged changes

When to Use Resources vs. Tools

Use resources when:

• Your client supports resource subscriptions and you want live updates
• You need a lightweight snapshot without additional parameters
• You are building a resource-aware UI on top of MCP

Use tools when:

• You need filtering, pagination, or custom options (e.g. git_log with author, since, or grep)
• You need a response format other than JSON (e.g. markdown)
• You are performing any write operation

Public Agent Discovery Endpoints

The documentation site publishes additional static discovery metadata for crawlers and AI agents:

Endpoint	Purpose
/.well-known/mcp.json	Compatibility pointer to MCP server card documents
/.well-known/api-catalog	High-level discovery links for docs and MCP metadata
/.well-known/mcp/server-card.json	Server metadata card describing transport and capabilities
/.well-known/mcp/server-cards.json	Multi-card discovery list containing server-card.json references
/.well-known/http-message-signatures-directory	Web Bot Auth JWKS for request-signing key discovery
/.well-known/agent-skills/index.json	Index of published agent skill files with digest metadata
/.well-known/agent-skills/git-mcp-overview/SKILL.md	Human-readable skill profile for git-mcp
/robots.txt	Crawl policy including AI/model crawler directives

The site also exposes a minimal WebMCP imperative API registration script at /webmcp.js that registers a read-only docs navigation tool when navigator.modelContext is available.

These files are intended for web discovery of project capabilities and do not replace MCP runtime discovery inside connected clients.

Static Hosting Constraints

This site is hosted as static content (GitHub Pages), so discovery metadata is static at build time:

• No dynamic capability negotiation or per-client responses
• No request signing or auth handshakes at /.well-known/*
• No runtime mutation of skill indexes or server cards

For runtime behavior, always trust live MCP server capabilities exposed by the connected transport/session.

For Web Bot Auth specifically, this static site publishes a discoverable JWKS directory, but does not itself send signed crawler requests. Request signing and Signature-Agent / Signature-Input headers must be implemented by any active bot/agent HTTP client that calls other sites.