Claude Code and other coding agents start every session from zero — you re-explain architecture, conventions, and open tasks. Session 47 repeats the same mistakes session 12 already fixed.
AXME Code is an MCP plugin that gives coding agents what CLAUDE.md cannot: writable memory, structured decisions with enforcement levels, hook-based safety (not prompt suggestions), and session handoff when context limits or vacations interrupt work.
Oracle captures stack and patterns from your repo. Decisions separate contracts from facts — required vs advisory. Feedback memories stop repeated errors; pattern memories reuse what worked. Guardrails block force-push, denied paths, and protected branches before tools run.
Code complements AXME Cloud (durable agent workflows) and AXME Mesh (fleet governance). Install once per repo, call axme_context at session start, and pick up mid-task without the forty-minute re-brief.
FIVE PILLARS
What AXME Code adds to Claude Code.
Beyond a static CLAUDE.md file.
Oracle
Auto-built project knowledge base.
Learn more →Decisions
Required vs advisory rules.
Learn more →Memory
Feedback + patterns that persist.
Learn more →Safety hooks
Enforced before execution.
Learn more →Handoff
Resume after context limit.
Learn more →Multi-repo
Per-repo storage, workspace root.
Learn more →Prompt rules vs hook enforcement
CLAUDE.md only
# CLAUDE.md NEVER git push --force # agent may still run it
AXME Code
# hook intercepts before run if denied_command: block # + decisions, memory, handoff
Session start
# Call at the start of every session axme_context(path="/path/to/your/repo") # Loads oracle, decisions, safety, memories
Related reading
Deeper dives from the AXME blog.
How I Fixed Claude Code's Amnesia (Actually)
Claude Code forgets your project every session. CLAUDE.md helps, but it doesn't scale. I built an MCP server plugin that gives Claude Code persistent memory, decisions, and safety guardrails.
Read post →Decisions Are Not Memories. Most Agent Memory Systems Confuse Them.
'I learned we use Postgres' and 'we decided Postgres over MySQL because of JSON support' are different data. One is a fact. The other is a contract. Mixing them is why agent memory drifts.
Read post →Your AI Coding Agent Keeps Running `git push --force`. Prompts Don't Fix This.
You put 'never force push' in CLAUDE.md. The agent force-pushed anyway. Here's why prompt-level safety rules fail and what tool-level enforcement actually looks like.
Read post →
From the blog
Ship your first durable agent — in under 10 minutes.
Free tier. No credit card. Self-host or hosted — your choice.