axme_contextLABSCO SUMMARY
Decisions carry enforcement levels, which is what turns a note into a rule.
A memory the agent may ignore is worth much less than a decision it must satisfy before reporting completion — that distinction is the design here. The safety hooks only exist where the harness supports them: in other clients you get the same knowledge base and no automatic enforcement, which the project states plainly.
What it is
A persistent memory and safety layer for coding agents. It builds a knowledge base from your project, serves it over MCP so any client can read and write it, and — in Claude Code — installs hooks that block dangerous commands before they execute.
What you get
- axme_context — the full project context at session start: stack, decisions, patterns, glossary and the previous session's handoff
- axme_oracle — the project map: structure, tech stack, coding patterns and glossary
- axme_decisions and axme_save_decision — architectural decisions with enforcement levels, read and written
- axme_memories and axme_save_memory — durable notes about the project
- axme_safety and axme_update_safety — the safety rules, and changes to them
- axme_backlog, axme_backlog_add, axme_backlog_update — a backlog the agent maintains
- axme_ask_question, axme_answer_question, axme_list_open_questions — the agent raises a question instead of guessing, and you answer it later
- axme_begin_close and axme_finalize_close — structured session handoff so the next session starts where this one stopped
- axme_status, axme_worklog, axme_workspace — current state, what was done, and multi-repo workspace handling
- Pre-execution hooks that intercept dangerous commands at the harness level, in Claude Code
Requirements
Node.js 20 or newer on PATH. Install as a Claude Code plugin from the community marketplace, as a Cursor extension, or as a standalone binary with the published install script; then run the setup command in each project. Setup scans the project, builds the knowledge base, installs the safety hooks and writes the MCP entry. The tools work in any MCP client; the hooks and the background auditor are Claude Code specific.
Setup effort
Follow the published steps — curl -fsSL https://raw.githubusercontent.com/AxmeAI/axme-code/main/install.sh | bash
USE CASES
01
Starting a session already knowing the project
Load the stack, decisions, patterns and the previous session's handoff in one call.
02
Not re-explaining an architectural rule
Save the decision once with its enforcement level and let later sessions read it back.
03
Ending a session so the next one continues it
Begin the close, review what is recorded, and finalise the handoff.
04
Getting an answer instead of a guess
Have the agent raise the ambiguity as a question and answer it when you are back.
05
Keeping a running list of what is left
Add to the backlog as things surface and update items as they are dealt with.
06
Working across several repositories
Set up the workspace once so the knowledge base spans them.
19 tools
Listed in project docs. Your client may show a different list.
axme_oracleaxme_decisionsaxme_memoriesaxme_save_decisionaxme_save_memoryaxme_safetyaxme_update_safetyaxme_backlogaxme_backlog_addaxme_backlog_updateaxme_statusaxme_worklogaxme_workspaceaxme_begin_closeaxme_finalize_closeaxme_ask_questionaxme_list_open_questionsaxme_answer_questionBEFORE YOU INSTALL
What to know first
Tools you get
19 tools named in the project's documentation.
Requires
axme-code on PATH
Where it runs
All 5 documented clients run the server on your machine.
INSTALLATION
Connect your client
Server setup
TERMINAL
curl -fsSL https://raw.githubusercontent.com/AxmeAI/axme-code/main/install.sh | bash
cd <YOUR_PROJECT>
axme-code setupMCP Clients
Claude Code
CLAUDE CODE · JSON
{
"mcpServers": {
"axme": {
"command": "axme-code",
"args": [
"serve"
]
}
}
}axme-code setup writes this entry into the project's .mcp.json for you, along with the safety hooks — you only add it by hand if you skipped setup.
Cursor, Windsurf, Claude Desktop, Cline
JSON
{
"mcpServers": {
"axme": {
"command": "axme-code",
"args": [
"serve"
]
}
}
}
