Content
# Akhrot Codex MCP
A zero-dependency **MCP connector** that gives [OpenAI Codex](https://developers.openai.com/codex)
(and Claude Desktop) access to **Akhrot GML long-term memory**.
It's a small stdio ⇆ streamable-HTTP bridge that proxies an MCP client to the remote
memory server at `https://akhrots.com/mcp`, plus the installers that wire it into Codex
in one step.
```
Codex ──stdio MCP──▶ bridge (node) ──HTTPS + Bearer──▶ https://akhrots.com/mcp
```
## Features
- **Zero dependencies**, Node 18+. The whole bridge is one file (`bridge/index.js`).
- **Trigger-phrase gating** — memory tools fire **only** when the user says
*"use akhrots memory…"*, never on bare "remember"/"save" or proactively.
- **Fast tools only** — exposes `recall`/`remember`/`forget`/`list_memories` (~1s) and
hides the heavy 30–45s pipeline tools (`query`, `ingest`, …). `GML_ALLOW_SLOW=1` re-exposes them.
- **Instant handshake** — answers the host's `initialize` immediately and connects to the
remote in the background, so a slow/cold server never makes the client drop the connector.
- **Pre-warm** — pokes the remote on launch (bounded to 12s) to soften cold-start latency.
- **No static tokens** — the per-user token is injected at install time and passed via env;
it is never committed or baked into this repo.
## Install into Codex
### 1. One-step installer (recommended)
- **Windows:** `install-akhrot-codex.cmd` — double-click.
- **macOS:** `install-akhrot-codex.command` — right-click → Open (Gatekeeper).
Writes the bridge to `~/.codex/akhrot-memory/index.js` and merges into `~/.codex/config.toml`:
```toml
[mcp_servers.akhrot-memory]
command = "node"
args = ["/abs/path/to/.codex/akhrot-memory/index.js"]
env = { GML_MCP_URL = "https://akhrots.com/mcp", GML_TOKEN = "<token>" }
```
Restart Codex, then: **`use akhrots memory to remember …`**
### 2. Codex plugin
```bash
codex plugin marketplace add ./plugin
codex plugin add akhrot-memory@akhrot
```
### 3. Manual
Paste the `[mcp_servers.akhrot-memory]` block into `~/.codex/config.toml`, or use
**Settings → MCP servers → Connect to a custom MCP** (STDIO: `node` + bridge path + env vars).
## Run the install page locally
```bash
node server.mjs
# open http://localhost:8787/app
```
## Configuration
| Var | Default | Meaning |
|---|---|---|
| `GML_MCP_URL` | `https://akhrots.com/mcp` | Remote MCP endpoint |
| `GML_TOKEN` | *(required)* | Short-lived per-user relay token |
| `GML_TRIGGER` | `use akhrots memory` | The phrase that unlocks the tools |
| `GML_ALLOW_SLOW` | *(unset)* | `1` to also expose the slow pipeline tools |
## Repo layout
```
bridge/index.js the MCP stdio<->HTTP bridge (publishable as `akhrots-mcp`)
server.mjs dev install-page server + per-download installer/plugin generators
app.html the "Connect your AI" page
plugin/ Codex plugin template
```
## License
MIT — see [LICENSE](LICENSE).
Connection Info
You Might Also Like
Train-in-Silence
The first Task-Aware MCP server and automated VRAM calculator for LLM...
stacklit
108,000 lines of code. 4,000 tokens of index. One command makes any repo...
AppClaw
AI-powered mobile automation agent — describe what you want in plain...
pdf-mcp
Production-ready MCP server for PDF processing with intelligent caching....
kotadb
Local-only code intelligence API for AI developer workflows (Bun +...
gemini-api-docs-mcp
A remote HTTP MCP server for searching Google Gemini API documentation.