Aura MCP — the Model Context Protocol server for git-native version control
Aura ships a 29-tool MCP server that lets Claude Code, Cursor, and any MCP client read AST graphs, log signed intent, snapshot files, and prove behaviour against your Git repo.
Aura MCP is the Model Context Protocol server that ships inside the aura CLI. It exposes 29 tools that any MCP client (Claude Code, Cursor, Zed, Continue, custom agents) can call to interrogate, mutate, and verify a Git-native semantic graph of your repository.
How to install
# 1. Install the Aura CLI (macOS / Linux)
curl -fsSL https://auravcs.com/install.sh | bash
# 2. Add Aura to your MCP client config
# Claude Desktop: ~/Library/Application Support/Claude/claude_desktop_config.json
{
"mcpServers": {
"aura": {
"command": "aura",
"args": ["mcp"]
}
}
}
# 3. Restart the client. Tools appear under aura_*.The 29 tools, grouped
- State & planning ·
aura_status,aura_plan_discover,aura_plan_lock,aura_plan_next,aura_prove. - Intent & provenance ·
aura_log_intent,aura_attest_verify,aura_attest_list,aura_handover. - Safety & recovery ·
aura_snapshot,aura_snapshot_list,aura_rewind,aura_pr_review,aura_suggest_edit,aura_doctor. - Live collaboration ·
aura_live_impacts,aura_live_resolve,aura_live_sync_push,aura_live_sync_pull,aura_live_sync_status. - Multi-agent + sentinel ·
aura_msg_send,aura_msg_list,aura_sentinel_inbox,aura_sentinel_send,aura_sentinel_status,aura_sentinel_agents,aura_sentinel_release,aura_orchestrate_status. - Memory & session ·
aura_memory_read,aura_memory_write,aura_memory_compact,aura_memory_forget,aura_session_resume,aura_session_summarize.
Why MCP and not a custom protocol
MCP is the agent-side equivalent of LSP for editors. By speaking it natively, Aura works with every current and future MCP-aware client without bespoke integrations. The agent doesn't need to know your repo is Aura-instrumented. It just calls aura_log_intent the same way it would call any other tool, and gets a signed block + sigstore receipt back.
Token-Oriented Object Notation (TOON)
MCP tool responses are encoded in TOON: Aura's compact serialisation format that uses 30–60% fewer tokens than equivalent JSON for structured data. That matters when an agent calls aura_pr_review on a 200-file diff and needs the response to fit inside a single context window.
What is MCP?
The Model Context Protocol — a standard interface for exposing tools to an AI client. A server declares what it can do, and any MCP-speaking client can call it. It is how an agent gets abilities beyond reading and writing files.
How do you add an MCP server to Claude Code?
Add an entry under mcpServers in the client config naming the command to run, then restart the client. The server’s tools then appear to the agent by name and are called like any other tool.
What can an MCP server do that a CLI cannot?
Be discovered. A CLI has to be described in a prompt and invoked through a shell; an MCP server advertises its tools, arguments and types to the client directly, so the agent knows what exists without being told and gets structured results back.
Aura Crew vs Claude Code Loops: the autonomous work-loop, compared
Claude Code defines four loop types — turn-based, goal, time, and proactive. Aura Crew is the same idea productized: a dependency-ordered work-loop with proof, collision-safety, and any agent. Full comparison.
Autonomous coding agent loops, explained: turn-based, goal, time, and proactive
A coding agent loop is an agent repeating cycles of work until a stop condition is met. The four types explained — turn-based, goal-based, time-based, proactive — with when to use each and how to keep quality high.
How to run multiple coding agents in parallel without merge conflicts
Running several AI coding agents at once collides on shared files. The fix: a worktree per agent, a live team radar, and soft-to-hard zone claims so parallel agents never touch the same symbol. How Aura Crew does it.
Goal-based agent loops: giving a coding agent a definition of done
A goal-based loop keeps a coding agent iterating until a success criterion is met. The trick is a deterministic definition of done — tests, a score, a proof — not an LLM guessing at "good enough." How to write one.