Production Ready · v2.4.0 · build 113

Your AI never forgets.

Self-hosted knowledge server for AI coding agents. 45 MCP tools, institutional memory, skill evolution. Works with claude.ai, Claude Code, Cursor, Windsurf, and any MCP client.

latest Build 113 — one-click Self-Update from the Admin UI, server-side write-safety guard, dev-login & 4-level ACL.
mxLore — claude code session

What happens after CLAUDE.md?

Every AI coding session starts from zero. Your decisions, your specs, your lessons — gone.

Without mxLore

  • Every session starts from scratch — AI has no memory
  • CLAUDE.md grows to 500+ lines, nobody reads it
  • Decisions are made, forgotten, then contradicted
  • Skills are static — no learning from mistakes
  • Each developer maintains their own context files
  • Cross-project knowledge? Copy-paste at best

With mxLore

  • AI recalls decisions, specs, and lessons automatically
  • Structured knowledge DB replaces bloated markdown
  • Decisions tracked as ADRs with full history
  • Skills evolve — auto-tuning based on real feedback
  • Shared team knowledge with per-developer access
  • Cross-project search, relations, and recall

Built for teams who ship, not demo.

Dev Teams

Shared knowledge that survives context switches. Each team runs their own server — one developer can work across multiple teams seamlessly. Stop re-explaining architecture to your AI.

CTOs & Architects

Architectural decisions tracked and enforced — from a chat window. Open claude.ai, ask “What did my team decide last week?” and get instant answers. No terminal needed. Data-driven skill evolution.

Product Managers

Write specs, review decisions, track progress — all from claude.ai. No terminal, no CLI. Everything you create is instantly available to your development team’s AI agents.

Regulated Industries

Self-hosted, GDPR-compliant, audit-ready. Your code knowledge never leaves your infrastructure. Full access logs and revision history.

Built for real workflows

Not a toy. Not a demo. 45 production tools forged through months of daily use across real projects.

45 Production MCP Tools
Search, create, update, recall, orchestrate. Every tool battle-tested in real codebases, handling real complexity.
mx_search, mx_recall, mx_create_doc ...
Institutional Memory
AI never forgets decisions, specs, or lessons. mx_recall retrieves context instantly with gate-matrix scoring and auto-learn.
mx_recall + recall_outcome
Semantic Search
Find documents by meaning, not just keywords. Provider-agnostic embedding API with MariaDB VECTOR. Hybrid search combines keyword + semantic results automatically.
VEC_DISTANCE_COSINE + hybrid scoring
Skill Evolution
12 skills that learn over time. Auto-tuning adjusts rules based on feedback. Skills get smarter with every session.
auto-tuning threshold: 10
/mxOrchestrate — One Command, Full Workflow
Your AI has a conductor. It coordinates plans, specs, decisions, code reviews, and health checks. Calls the right skill at the right time. Add your own Superpowers — the orchestrator picks them up automatically.
mxPlan · mxSpec · mxDecision · mxDesignChecker · mxBugChecker · mxHealth · mxSave · + your skills
Self-Hosted
Your server, your data, your rules. No cloud dependency, full GDPR compliance. Single binary, no runtime dependencies. Just download and run.
zero external dependencies
Encrypted Project Migration
Ship a full project — docs, specs, plans, decisions, ACL — to another mxLore instance as a single encrypted .mxbundle file. AES-256-GCM via Windows CNG + PBKDF2 key-derivation. 5-step wizard in the Admin UI covers conflict resolution, developer re-mapping, and audit.
contractor hand-off · air-gapped · cross-instance
API Key Lifecycle
Keys rotate on schedule, warn before expiry, offer grace periods, and revoke cleanly. RFC 7807 application/problem+json surfaces every reason — no more stale keys silently failing in production.
rotate · warn · grace · revoke
Dev-Login & 4-Level ACL
Non-admin developers log into the Admin UI with their own keys. Four access levels — read, comment, read-write, admin-bypass — gate 20+ server routes. Project creators auto-grant themselves read-write. Shipped build 110, 2026-04-24.
read · comment · read-write · admin-bypass
One-Click Self-Update new
Upgrade your mxLore server directly from the Admin UI — the server pulls the latest release from GitHub, swaps the binary, and restarts itself. User-config INIs are preserved by an explicit skiplist. Marker-recovery survives interrupted updates.
github releases · ini-preserving · marker-recovery

One developer. Multiple teams. Zero overlap.

Each project connects to its own mxLore server through a dedicated proxy. Teams stay autonomous — different databases, different API keys, different knowledge. One developer can seamlessly work across all of them.

Project A → Proxy A →
Server A ERP Team · own DB · own API key
Project B → Proxy B →
Server B Mobile Team · own DB · own API key
Project C → Default →
Global Server no .mcp.json — uses global config

Dev Teams

Per-team server. One developer switches context automatically via per-project proxy.

Agencies

Client A's knowledge never leaks into Client B. Full isolation, one laptop.

Enterprise

Department-level silos by design. Cross-team sharing via opt-in project relations.

How does mxLore compare?

From flat files to full knowledge management.

CapabilityCLAUDE.mdMemory MCPmxLore
Persistent knowledgeFile onlyKey-valueStructured DB
Cross-project searchNoNoFull-text + relations
Multi-developerNoNoACL + per-dev keys
Decision tracking (ADRs)ManualNoFirst-class doc type
Skill evolutionNoNoAuto-tuning + metrics
Institutional memoryNoBasicRecall + gate levels
Session orchestrationNoNoWorkflows + team agents
AI batch maintenanceNoNo12 boot-time jobs
Admin UINoNoTabbed dashboard, INI editor, intelligence
Encrypted project migrationPlain JSON or nonePlain JSON or noneAES-256-GCM .mxbundle + PBKDF2
API key lifecycleNo lifecycle, manualNo lifecycle, manualRotation, expiry warnings, grace, revoke
RFC 7807 error formatOpaque 500Opaque 500application/problem+json on 5xx
Multi-server / Multi-teamNoNoPer-project server
Works in chat UI (claude.ai)NoNoFull MCP access
Self-hostedLocal fileVariesYour server
The Core Idea

your whole team shares onebrain.

whether they code or chat
claude.ai Claude Code Cursor Windsurf Cline

One knowledge base. Two interfaces.

Your CTO reviews decisions in chat. Your developer codes with full context. Same memory, different window.

claude.ai
CTO
“What did we decide about caching?”
mx_recall found ADR-0034 + 3 lessons
“Event-driven invalidation, confirmed in Session 142”
PM
“Create a spec for the payment redesign”
mx_create_doc Spec #1087 saved
5 acceptance criteria defined
Same server.
Same database.
Claude Code
Developer
> /mxOrchestrate resume
  3 docs changed since last session
  ✔ New spec from CTO: Payment Redesign (#1087)
> /mxPlan "Implement per Spec #1087"
  ✔ Recalled: ADR-0034
  ✔ Plan #1088 — 6 milestones
  ✔ Linked to Spec #1087 (created in claude.ai)

CTOs & PMs

Review decisions from a chat window. No terminal needed.

Developers

Code with full context from yesterday’s discussions.

Consultants

Separate client knowledge bases, one chat window for all.

Production metrics, not marketing

Every number from actual deployment. No inflated benchmarks.

45
MCP Tools
search, recall, batch, agents
12
Skills
self-evolving with auto-tuning
<50ms
Response Time
p95 tool call latency
12
Batch Jobs
automated AI boot tasks

Watch a complete session

One command. Five skills. Zero manual steps.

mxLore — workflow demo

Three steps. Fifteen seconds.

From zero to institutional memory. No JSON editing, no config wikis.

1. Deploy Server
Binary + MariaDB. Schema auto-creates.
Download here
2. Open Admin UI
localhost:8081 → new member → copy API key
3. Connect Your AI
Run /mxSetup. Paste your API key.
4. AI Remembers
Every decision, spec, and lesson persisted.
Setup (once per developer)
# 1. Download the setup skill (one-liner)
mkdir -p ~/.claude/skills/mxSetup
curl -o ~/.claude/skills/mxSetup/SKILL.md https://raw.githubusercontent.com/MicrotronX/mxLore-skills/main/mxSetup/SKILL.md

# 2. Open the Admin UI in your browser — this is where the API key lives
#    http://localhost:8081   (first start has no login)
#    Click "New member" → enter a name → copy the generated mxk_... key

# 3. Back in Claude Code: run mxSetup with that key
/mxSetup mxk_your_api_key_here

# mxSetup asks: Server URL? → http://localhost:8080/mcp
# Downloads skills from GitHub, installs proxy, configures MCP
# Done in ~15 seconds. 45 tools ready.
Setup for claude.ai (Web / Mobile / Desktop)
# No CLI. No terminal. Just chat.

# 1. Get an API key — two ways:
#    (a) You host mxLore yourself → open http://localhost:8081
#        → click "New member" → copy the generated mxk_... key
#    (b) Someone else hosts it → ask your team admin for a key

# 2. Open claude.ai → Settings → Integrations → Add MCP Server
Name:        mxLore
Server URL:  http://your-server:8080/mcp
Auth Header: Authorization: Bearer mxk_your_api_key

# 3. Done. All 45 MCP tools available in any chat.

# Try it:
you: "Search my project knowledge about authentication"
# → Claude automatically uses mx_search

you: "Create a new spec for payment redesign"
# → Claude creates the doc via mx_create_doc

# Note: Skills (/mxOrchestrate, /mxPlan etc.) are Claude Code only.
# claude.ai gets all 45 MCP tools directly.
First session — using skills, not raw calls
# Start your session — orchestrator loads context automatically
/mxOrchestrate resume

# Plan a feature — spec, plan, and decisions tracked in DB
/mxPlan "Add user authentication with OAuth2"

# AI recalls relevant decisions automatically
you: "How did we handle token refresh last time?"
# → mx_recall finds ADR-0042 + 3 related lessons

# Review your code — skills learn from your feedback
/mxDesignChecker
/mxBugChecker

# Save everything — session notes, state, lessons extracted
/mxSave

Simple pricing. No subscriptions.

One-time purchase. All features. All updates. Forever yours.

Free
€0
Personal projects, evaluation, open source, students.
  • All 45 MCP tools
  • All 12 skills
  • Unlimited projects
  • Full source code access
  • Community support (GitHub Issues)
Commercial
€399 €149
Any size team. Unlimited everything. One-time payment, use forever.
  • Everything in Free
  • Commercial use — freelancer to enterprise
  • Unlimited servers, developers & projects
  • All future updates included
  • Priority email support

Source available · Server: BSL 1.1 (converts to Apache 2.0) · Skills: MIT · Full details →

Frequently asked questions

Yes! Add your MCP server in claude.ai Settings → Integrations. All 45 tools work — search knowledge, create docs, recall decisions. Your CTO can review architecture in a chat window while developers code with the same knowledge in Claude Code. Available on Claude Pro, Team, and Enterprise plans.
Free for personal and non-commercial use. Commercial license: €149 one-time (Early Adopter, regular €399). Any size team, unlimited servers, unlimited developers. All updates included. No subscriptions, no seat counting. Contact info@microtronx.com for licensing.
Never. mxLore is fully self-hosted. Your data stays on your infrastructure. The only external call is the optional AI Batch feature which uses your own Claude API key — and even that can be disabled.
CLAUDE.md is great for project rules but doesn't scale. mxLore adds structured storage, cross-project search, decision tracking, skill evolution, multi-developer access, and institutional memory that persists across sessions.
Yes. mxLore speaks standard MCP (Model Context Protocol). Any MCP-compatible client can connect — Claude Code, Cursor, Windsurf, Cline, Copilot CLI, and others.
Currently Windows only (x64). Linux/Docker support is planned for a future release. The MCP protocol is platform-independent — only the server binary is Windows-specific.

Ready to give your AI a memory?

Deploy in minutes. 15-second onboarding. Your infrastructure, your data, your rules.