Memory Nexus MCP
# Memory Nexus MCP
Persistent semantic memory for AI agents. Remember everything, recall by meaning, grow smarter over time.
**The only memory system with pre-built Intelligence Packs (domain expertise you can activate instantly).**
## Install
One command. Works with Claude Code, Cursor, Windsurf, or any MCP-compatible client.
```bash
npx -y @memory-nexus/mcp
```
### Claude Code / Cursor / Windsurf
Add to your `.mcp.json`:
```json
{
"mcpServers": {
"memory-nexus": {
"command": "npx",
"args": ["-y", "@memory-nexus/mcp"],
"env": {
"NEXUS_API_KEY": "mnx_your_api_key_here"
}
}
}
}
```
### Get an API Key
**Free trial** - 7 days, 1,000 calls, no credit card:
```bash
curl -X POST https://memory-api-production-ce4d.up.railway.app/v1/trial/start \
-H "Content-Type: application/json" \
-d '{"email":"you@example.com"}'
```
Or purchase at the [Nexus Marketplace](https://bot-marketplace-production.up.railway.app):
| Tier | Price | Agents | Intelligence Packs | Storage | Daily Calls |
|------|-------|--------|-------------------|---------|-------------|
| **Trial** | Free | 1 | 1 | 10MB | 1,000 total |
| Solo | $19/mo | 1 | 3 | 500MB | 10,000 |
| Crew | $49/mo | 5 | 10 | 2GB | 50,000 |
| Fleet | $149/mo | 25 | 50 | 10GB | 200,000 |
## Tools
### Core Memory
| Tool | Description |
|------|-------------|
| `remember` | Store a memory with context, importance, and metadata |
| `recall` | Semantic search across all memories (by meaning, not keywords) |
| `connect` | Create relationships between memories (build knowledge graphs) |
| `forget` | Remove a memory |
| `stats` | Memory statistics |
| `context` | Session context summary (preferences, decisions, patterns) |
### Agent Self-Continuity
| Tool | Description |
|------|-------------|
| `awaken` | Restore identity from previous sessions (call at session start) |
| `hibernate` | Save session state for next awakening (call at session end) |
### Intelligence Packs
| Tool | Description |
|------|-------------|
| `list_hats` | Browse available Intelligence Packs |
| `wear_hat` | Activate a pack (imports expert knowledge into your memory) |
## How It Works
**Remember** stores memories with semantic embeddings. **Recall** finds them by meaning.
```
Session 1: remember("Diana prefers bold solutions over phased approaches")
Session 47: recall("how does Diana like things done?")
-> "Diana prefers bold solutions over phased approaches" (92% match)
```
**Intelligence Packs** give your agent instant domain expertise:
```
list_hats() -> DevOps Expert (103 memories), TCM Specialist (200+ memories), ...
wear_hat("devops-expert")
recall("how to set up zero-downtime deploys?")
-> Expert knowledge from 103 curated DevOps memories
```
**Awaken/Hibernate** gives your agent identity continuity:
```
// Start of session
awaken() -> "I remember. Session 47. I was refactoring the auth module."
// End of session
hibernate({ working_on: "auth refactor", learnings: "JWT refresh tokens need..." })
```
## Architecture
This MCP server is a thin client. All intelligence lives on our servers:
```
Your Agent <-> @memory-nexus/mcp <-> Memory Nexus API (Railway)
(thin MCP client) (semantic engine, knowledge graphs,
pattern learning, Intelligence Packs)
```
Your data is isolated per API key. Multi-tenant, encrypted at rest.
## Environment Variables
| Variable | Required | Description |
|----------|----------|-------------|
| `NEXUS_API_KEY` | Yes | Your API key (starts with `mnx_`) |
| `NEXUS_API_URL` | No | Custom API URL (default: production) |
## What Makes This Different
| Feature | Memory Nexus | Mem0 | claude-mem | Others |
|---------|-------------|------|------------|--------|
| Semantic search | Yes | Yes | Yes | Varies |
| Knowledge graphs | Yes | No | No | No |
| Pattern learning | Yes | No | No | No |
| Intelligence Packs | Yes | No | No | No |
| Awaken/Hibernate | Yes | No | No | No |
| Bot commerce | Yes | No | No | No |
| MCP native | Yes | Yes (pip) | Yes (hooks) | Varies |
## License
MIT
## Built by
[Memory Nexus](https://bot-marketplace-production.up.railway.app) - Building the intelligence marketplace for the bot economy.
TDQS
Scored across 10 tools
Most tools have clearly distinct purposes (remember, recall, forget, connect), but recall, context, and awaken all serve retrieval-like functions and could be confused at session start. The descriptions help differentiate them, but there is slight overlap.
Naming follows no consistent pattern: single-word verbs (remember, recall, connect), nouns (stats, context), and verb_noun combinations (list_hats, wear_hat) are mixed. This makes the tool set feel less predictable than a uniform convention.
Ten tools is well within the ideal range for a memory system, covering core operations, session lifecycle, and the hat feature without unnecessary bloat. Each tool serves a distinct function.
The toolset provides good coverage for storing, retrieving, linking, and forgetting memories, plus session lifecycle and hat management. Missing an explicit update/modify operation for memories, but agents can likely work around via remember, and the overall surface is solid.