Open-Source Chat Clients Setup — Docs

Overview

The open-source ecosystem has produced a wave of capable chat clients — Open WebUI, LibreChat, Cherry Studio, Jan, AnythingLLM, Msty, and Goose among them. They run whatever model you point them at (local via Ollama or llama.cpp, or any hosted API), and most now speak MCP. Connecting Tempreon gives all of them the same thing: your AI loads your Core Imprint, searches your Knowledge Vault, and remembers what it learns — across every tool.

Every client on this page uses the same server URL, also shown in your Bridges dashboard:

https://api.tempreon.com/functions/v1/tempreon-mcp/mcp

The mcp-remote proxy. Clients without a built-in OAuth flow use mcp-remote — a local stdio bridge that opens the browser OAuth for them and forwards tool calls to Tempreon. Wherever a client below needs it, this is the command (requires Node 20+):

npx -y mcp-remote https://api.tempreon.com/functions/v1/tempreon-mcp/mcp

Prerequisites

A Tempreon account with at least one available Bridge slot
One of the clients below, updated to a current build
For the mcp-remote clients: Node.js 20 or later installed

Clients with native OAuth — connect directly

Open WebUI

Self-hosted, browser-based, works with any model backend (Ollama, OpenAI-compatible APIs, OpenRouter). Native MCP support — Streamable HTTP only — arrived in v0.6.31.

Open Admin Panel → Settings → External Tools
Add a new tool with the type MCP (Streamable HTTP)
Paste the server URL and set authentication to OAuth 2.1
Each user completes the browser OAuth individually the first time they use it

Two gotchas: if you run Open WebUI in Docker, set a fixed WEBUI_SECRET_KEY — without it, stored OAuth tokens break every time the container restarts. And set your model's Function Calling setting to Native, or tools won't be called reliably.

LibreChat

Self-hosted chat platform for any model provider. Add Tempreon to librechat.yaml and restart:

mcpServers:
  tempreon:
    type: streamable-http
    url: https://api.tempreon.com/functions/v1/tempreon-mcp/mcp

When no credentials are configured, dynamic client registration is the default — exactly what Tempreon expects. Users trigger the browser OAuth from the chat UI the first time they use the server.

Goose

Block's open-source AI agent — desktop and CLI, any model provider.

One-click (Desktop): Add to Goose — with Goose Desktop installed, this opens the add-extension dialog pre-filled
Desktop (manual): Settings → Extensions → Add custom extension → type Streamable HTTP → name + URL
CLI, one session: goose session --with-streamable-http-extension "https://api.tempreon.com/functions/v1/tempreon-mcp/mcp"
CLI, persistent: add to config.yaml:

extensions:
  tempreon:
    enabled: true
    type: streamable_http
    name: Tempreon
    uri: https://api.tempreon.com/functions/v1/tempreon-mcp/mcp
    timeout: 300

Goose starts the browser OAuth automatically when the server asks for it. On current builds it may re-prompt for authorization each session — approve and continue.

Msty

Desktop app for local and hosted models; its tool system is called the Toolbox, available on the free desktop app.

Open the Toolbox and add a tool
Choose HTTP and paste the server URL
Pick OAuth authentication — the browser flow opens when the connection is first used

Msty supports Streamable HTTP only; SSE servers aren't supported. Tempreon is Streamable HTTP, so it connects directly.

Cherry Studio

Desktop app (Windows, macOS, Linux) with 20+ model providers, including Ollama, OpenRouter, Qwen, and DeepSeek.

Open Settings → MCP Servers → + Add Server
Set the type to streamableHttp and paste the server URL
Save, then enable Tempreon via the MCP tools icon under the chat input

Cherry Studio ships an OAuth flow, but it has been uneven across releases — test it, and if authorization fails, switch to a stdio entry using the mcp-remote command from the callout above. Keep the app updated; recent releases also carry security fixes.

Clients that use the mcp-remote proxy

Jan

Free, open-source desktop app running local models via llama.cpp, plus cloud providers. Jan's HTTP transport has no OAuth flow, so use the proxy:

Open Settings → MCP Servers and toggle the MCP host on
Click + Add MCP Server
Configure a stdio server:
- Command: npx
- Args: -y, mcp-remote, https://api.tempreon.com/functions/v1/tempreon-mcp/mcp

The first run opens your browser for the Tempreon OAuth; after that the proxy reuses its stored tokens.

AnythingLLM

Desktop and Docker workspace app for any model. MCP servers are configured in anythingllm_mcp_servers.json in the storage plugins/ directory:

{
  "mcpServers": {
    "tempreon": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://api.tempreon.com/functions/v1/tempreon-mcp/mcp"]
    }
  }
}

If you go direct instead, AnythingLLM's HTTP type keyword is "type": "streamable" (omit it and SSE is assumed) — but its authentication is headers-only with no OAuth, so for Tempreon use the mcp-remote stdio entry above. Refresh the server list on the Agent Skills page, and invoke tools via @agent.

Two more you might be using

Continue.dev still works locally with type: streamable-http in ~/.continue/config.yaml and a native OAuth browser prompt, but it is no longer maintained following its acquisition.
Former Roo Code users: Roo shut down in May 2026 — its community fork continues with the same config format, and needs the mcp-remote proxy.

Troubleshooting

The connection fails immediately. Confirm the URL ends in /mcp with no trailing slash: https://api.tempreon.com/functions/v1/tempreon-mcp/mcp.

npx: command not found, or mcp-remote won't start. The proxy needs Node.js 20 or later — install or update Node, then retry.

Tools connect but the model never calls them. Check the model side: it needs tool-calling support, and in Open WebUI the model's Function Calling setting must be Native. Smaller local models can also struggle with large tool lists.

OAuth tokens stop working after a restart. In Dockerized Open WebUI, set a fixed WEBUI_SECRET_KEY. Elsewhere, remove and re-add the server to run a fresh OAuth flow.

Also connect: LM Studio · Connect Any MCP Client · Claude Desktop

Try saying...

"Start my session." — triggers session_start to load your Core Imprint
"What do you know about me from Tempreon?" — a quick end-to-end test
"Remember that [insight] for next time." — proves the write path works from your client