Rediscovery every prompt
Dozens of file reads per task, repeated each conversation, billed in tokens.
For teams building with AI agents
Give agents the repo context they keep guessing at.
Agents re-crawl and guess at conventions every prompt. That burns tokens and shows up in review. Moxie serves conventions, docs, gaps, and commands over MCP - scoped and current.
MCP session
read-only→ get_conventions()
12 rules · paths cited
→search_docs("webhook")
top hit: billing webhook behavior
ready to write
Two scoped lookups, not 40 file reads
The problem
Sound familiar?
Conventions ignored
Agents guess patterns you already standardized - you fix it in review.
Wrong files in context
Stuffing the repo is expensive and still misses house rules.
The fix
What agents get over MCP
Scoped lookups, not 40 reads
Conventions and docs on demand - structured and source-cited.
Right the first time
House rules and boundaries up front, not inferred from a partial crawl.
Docs with the change
Matching doc updates can land in the same PR as the code work.
FAQ
Questions, answered
Which agents and tools does Moxie support?+
Any MCP-aware client. Claude Code, Cursor, and Codex can query Moxie's Model Context Protocol server directly for conventions, docs, gaps, and verified commands.
Is the MCP server read-only?+
Yes. It's context-only and scoped per repository: conventions, docs, and citations, never billing, settings, or the merge button.
How much does this actually save?+
One scoped lookup replaces dozens of exploratory file reads per task, repeated across every prompt and conversation. You spend fewer tokens rediscovering the repo, and redo fewer wrong guesses in review.
Get started
See what the first index finds.
Connect a repository and review your first source-cited docs in minutes. 14-day trial, no charge today, cancel anytime.