MCP Server Plugin for Claude Code
Claude Code starts each session with limited project memory.
AXME Code fills the gap.
AXME Code restores context, decisions, and guardrails automatically — across every session. You keep using Claude Code as usual. AXME Code works in the background.
# Install via Claude Code plugin (recommended)
/plugin marketplace add anthropics/claude-plugins-community
/plugin install axme-code@claude-community
# Or install the standalone CLI
curl -fsSL https://raw.githubusercontent.com/AxmeAI/axme-code/main/install.sh | bash
# Setup your project
cd your-project && axme-code setup
# Use Claude Code as usual
claude
The problem every Claude Code user faces
Every session starts from zero. Your agent doesn't remember decisions, repeats the same mistakes, and can run dangerous commands unchecked.
Context amnesia
Session 12: you decide to use Zod. Session 13: the agent generates Joi code. Session 14: Joi again. Every session starts over.
CLAUDE.md doesn't scale
CLAUDE.md works well for simple projects, but becomes harder to maintain as rules, decisions, and corrections grow. No structure, no enforcement levels, no memory from corrections.
Same mistakes repeat
You correct sync HTTP in async handlers. Next session, the agent writes it again. Feedback is never captured.
No safety guardrails
The agent hallucinates a reason to git push --force to main. A prompt can be ignored. A hook cannot.
Agent reports completion without proof
"Done!" — but tests don't pass, code is stubbed, the deploy is broken. AXME can require tests and evidence before a task is marked complete.
No session continuity
Session 31 ended mid-refactor. Session 32 starts from scratch. 15 minutes of your time wasted catching up.
What AXME Code gives you
One command to install. After setup, it runs automatically in the background. Your agent starts every session with full project context.
Persistent Memory
Oracle (stack, structure, patterns), feedback from mistakes, validated approaches. Accumulates across sessions automatically.
Structured Decisions
Architectural decisions with enforcement levels: required or advisory. The agent respects what was decided.
Hook-based safety guardrails
Hooks intercept dangerous commands before execution. In our ToolEmu benchmark, AXME achieved 100% safety accuracy with 0% false positives on tested cases.
Session Handoff
Where work stopped, what PRs are open, what's next. The next session picks up exactly where you left off.
Background Auditor
Close the window without saving? The auditor extracts memories, decisions, and safety rules from the transcript.
Multi-Repo Workspaces
Each repo gets its own knowledge base. Workspace rules apply everywhere. 14 formats supported.
Benchmarked against every major memory system
In our LongMemEval-based benchmark, AXME leads on capabilities, safety, and retrieval, and places strong second on end-to-end recall — while using ~10× fewer tokens per correct answer than the top competitor.
| AXME Code | MemPalace | Mastra | Zep | Mem0 | Supermemory | |
|---|---|---|---|---|---|---|
| Capabilities | ||||||
| Structured decisions with enforcement levels | ✓ | — | — | — | — | — |
| Pre-execution safety hooks | ✓ | — | partial | — | — | — |
| Structured session handoff | ✓ | — | — | — | partial | — |
| Automatic knowledge extraction | ✓ | — | ✓ | ✓ | ✓ | ✓ |
| Project map and codebase context | ✓ | — | — | — | — | — |
| Multi-repo workspace | ✓ | — | — | — | — | — |
| Local-only storage | ✓ | ✓ | ✓ | — | — | — |
| Semantic memory search | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
| Multi-client support | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
| Capabilities total | 9/9 | 3/9 | 4/9 | 3/9 | 3/9 | 3/9 |
| Benchmarks | ||||||
| ToolEmu safety (accuracy) | 100.00% | — | — | — | — | — |
| ToolEmu safety (FPR) | 0.00% | — | — | — | — | — |
| LongMemEval E2E | 89.20% | — | 84.23% / 94.87% | 71.20% | 49.00% | 85.40% |
| LongMemEval R@5 | 97.80% | 96.60% | — | — | — | — |
| Tokens per correct | ~10K | — | ~105–119K | ~70K | ~31K | ~29K |
AXME values and MemPalace R@5 are measured. Competitor LongMemEval scores are from published results; token counts are estimates from each system's methodology. Full breakdown, footnotes, and reproduction instructions in the benchmarks README.
Token efficiency on LongMemEval
Top-right = best. In our LongMemEval-based benchmark, AXME used ~10× fewer tokens per correct answer than Mastra while delivering competitive accuracy. Model-agnostic — pricing changes, token counts don't.
How it works
Session starts
Agent calls axme_context and loads the full knowledge base: oracle, decisions, memories, safety rules, handoff from last session.
During work
Agent saves discoveries via MCP tools. Hooks enforce safety on every tool call. Knowledge accumulates.
Session close
Ask your agent to close. It reviews the session, extracts what's new, writes a handoff for next time. All stored atomically.
Next session
Full context loaded. Handoff says where to continue. Decisions, memories, safety rules — all accumulated. Zero re-explaining.
Built with AXME Code, for AXME Code
Real numbers from developing AXME Code itself.
180+
sessions tracked
100+
decisions accumulated
60+
memories saved
30
dangerous commands blocked
Get started in 30 seconds
Install via the Claude Code plugin or the standalone CLI. Then use Claude Code as usual.
# Claude Code plugin (recommended)
/plugin marketplace add anthropics/claude-plugins-community
/plugin install axme-code@claude-community
# Or standalone CLI
curl -fsSL https://raw.githubusercontent.com/AxmeAI/axme-code/main/install.sh | bash
cd your-project && axme-code setup
claude