ModelRein (formerly LMCP) is an HTTP protocol and reference server for controlling open-source foundation models: 27 primitives and 6 architecture adapters. Today it is used from the Rescot desktop app, which connects to your own GPU server over SSH (v0.4).
Rescot: Settings → Connections → LMCP connections, paste your SSH command and password; it forwards to local port 8030–8035, password in memory only (1.10.2).
| Fact | Value | Type | As of | Source |
|---|---|---|---|---|
| How Rescot reaches your GPU server | paste the SSH command + password; the built-in SSH client forwards to one local port in 8030–8035 | Verified by us | 2026-10-22 | Rescot 1.10.2 source |
| How the SSH password is kept | memory only, 0 writes to disk, never sent to us | Verified by us | 2026-10-23 | Rescot privacy policy |
| Where ModelRein is set up in Rescot | Settings → Connections → LMCP connections: a Local card (detects NVIDIA GPUs) and a Connect-to-server card (SSH command + password); each server has a 'Used in Task mode' switch | Verified by us | 2026-11-04 | Rescot 1.10.3 source |
No: ModelRein v0.4 is not on PyPI (404 on 2026-11-04) and has no public repo. Today Rescot connects to your server over SSH and starts the service on port 8030.
| Fact | Value | Type | As of | Source |
|---|---|---|---|---|
| PyPI release status | not published: pypi.org returns 404 for lmcp and there is no public repository; there is no pip install route | Verified by us | 2026-11-04 | pypi.org |
| Default command Rescot runs to start ModelRein on the server | runs python3 -m uvicorn server:app --host 127.0.0.1 --port 8030 in ~/lmcp in the background, logging to ~/lmcp.log; the command is editable | Verified by us | 2026-11-04 | Rescot 1.10.3 source |
| Requirements | Python 3.10+; Linux / macOS 13+ / Windows 11; GPU optional; at least 8 GB RAM | Site docs | 2026-10-09 | lmcp-quickstart.html |
| Current version | v0.4 | Site docs | 2026-10-09 | about.html |
Rescot 1.10.2 ships 24 lmcp_ tools (e.g. lmcp_generate, lmcp_pipeline), on once a server is connected and 'Used in Task mode' is on (2026-10-22).
| Fact | Value | Type | As of | Source |
|---|---|---|---|---|
| ModelRein tools built into Rescot | 24 lmcp_ tools (including introspect, generate, compute_vector_auto, head_mask, layer_route, pipeline) | Verified by us | 2026-10-22 | Rescot 1.10.2 source |
| Tools exposed by the MCP bridge | 20 (the MCP bridge is not publicly distributed yet) | Site docs | 2026-10-09 | lmcp-quickstart.html |
| Where ModelRein is set up in Rescot | Settings → Connections → LMCP connections: a Local card (detects NVIDIA GPUs) and a Connect-to-server card (SSH command + password); each server has a 'Used in Task mode' switch | Verified by us | 2026-11-04 | Rescot 1.10.3 source |