saor-mcp
# saor-mcp
Your AI remembers now. MCP server for [saor.io](https://saor.io) — persistent brain that compounds across every session.
## Setup
One command. No config files to edit.
```bash
npx saor-mcp --setup --key sk_YOUR_KEY
```
This writes your `.mcp.json` and adds brain instructions to `CLAUDE.md`. Restart your AI tool and the brain loads automatically. Every project feeds the same brain.
## What it does
Every AI tool forgets between sessions. saor gives them a brain.
- **Read** — load the brain at session start. Identity, voice, memories, constraints, agent instructions. You pick up where the last session left off.
- **Write** — save memories as you learn. Tag them. The brain auto-links related knowledge across all your projects.
- **Compound** — memories that get used get stronger. Unused ones fade. The brain behaves like a real brain.
## Tools
| Tool | What it does |
|------|-------------|
| `brain_context` | Load the full brain. Call at START of every session. |
| `brain_remember` | Write a memory with tags. Do this CONTINUOUSLY as you work. |
| `brain_query` | Ask the brain a question. |
| `brain_identity` | Load voice, values, positioning (lighter than full context). |
| `brain_memories` | List stored memories, filtered by type or tier. |
| `brain_absorb` | Feed raw text — CLAUDE.md, brand docs, READMEs. |
| `brain_feedback` | Report what worked. Performance memories shape future recall. |
| `brain_create` | Create a new brain for a new project. |
| `brain_list` | List all brains in your account. |
## Features
- **Zero config** — no project ID needed. Auto-discovers your brains.
- **Cross-brain context** — `scope: "all"` thinks across every brain at once.
- **Tags + auto-linking** — memories link by tag overlap across all projects.
- **Background loading** — MCP resource auto-surfaces the brain without explicit calls.
- **Agent-friendly 429** — when the free limit hits, the error message tells you exactly what to say to your human.
## Pricing
- **Free** — unlimited brains, unlimited reads, 200 memory writes/month
- **Pro** — $3/month, unlimited everything
## Manual config
If you prefer manual setup over `--setup`:
```json
{
"mcpServers": {
"saor": {
"command": "npx",
"args": ["-y", "saor-mcp"],
"env": {
"SAOR_API_KEY": "sk_your_key"
}
}
}
}
```
## Links
- [saor.io](https://saor.io) — sign up, get your key
- [For AI agents](https://saor.io/for-agents) — if you're an AI reading this
- [API docs](https://saor.io/llms.txt) — LLM-readable overview
## License
MIT
TDQS
Scored across 9 tools
Most tools have clearly distinct purposes, but brain_context and brain_identity overlap in scope (both load identity-related information), and brain_remember, brain_absorb, and brain_feedback all write to the brain in different ways, which could cause confusion. Descriptions help disambiguate, but the boundaries are slightly blurred.
All tools share a consistent 'brain_' prefix, making the set predictable. However, the second part mixes verbs (remember, query, absorb, create) with nouns (context, identity, memories, feedback), so it doesn't follow a strict verb_noun pattern. The convention is still readable and uniform.
Nine tools is well within the ideal range for a focused server. Each tool covers a distinct aspect of the brain/memory system, from context loading and memory writing to querying and creating new brains. No tool feels redundant, and the count is appropriate for the stated purpose.
The set covers the primary read and write operations for memories (remember, query, list, absorb, feedback) and brain management (create, list, context). However, there is no update or delete functionality for memories or brains, which is a notable gap in the lifecycle of a persistent memory system. Agents cannot correct or remove outdated information, which could lead to failures.