Connect Any MCP Client — Tempreon Docs

Overview

Tempreon connects through the Model Context Protocol (MCP) — an open standard any compatible AI client can speak. If your client supports remote MCP servers with OAuth, it can connect to Tempreon.

This page walks through the universal setup flow. The pattern is the same across every OAuth-capable client, even if the menus and labels differ. If you're using a client we've fully tested (Claude Desktop, Claude Code, or ChatGPT), the specific guide will be faster. Use this page when you're on a newer or less common client, or when you want to understand what's actually happening under the hood.

When to use this page

Use this page if you're connecting Tempreon from:

A client that isn't covered by our specific guides (Cursor, Windsurf, Cline, Zed, a custom app built on an agent SDK, or any new client that speaks MCP)
A client you just started testing and want to verify the flow before anything else
A specific guide that didn't match your version's UI — the universal pattern here still applies

If you're on a client we've tested end-to-end, the specific guide will be faster. See When to use a client-specific guide below.

Prerequisites

Before connecting, you need:

A Tempreon account with at least one available Bridge slot (check your Bridges dashboard)
An MCP client that supports remote servers over Streamable HTTP with OAuth authentication
A browser reachable by your client — the authorization step happens in your default browser, not in the client itself

If you're not sure whether your client supports remote MCP servers, check its documentation for "MCP," "connectors," "custom apps," or "external tools." Clients that only support local (stdio) MCP servers through a subprocess can't connect to Tempreon directly without a shim — see Troubleshooting below.

The universal connection flow

Every OAuth-capable MCP client follows the same six-step pattern. Menu names and button labels vary. The flow does not.

1. Find your MCP server settings

Look in your client for a section called MCP Servers, Connectors, Custom Apps, Tools, or Integrations. It's usually under Settings. If your client has a command palette, searching for "MCP" will almost always find it.

2. Add a new server

Choose "Add custom server," "Add connector," or the equivalent. You'll be prompted for a name and a URL.

Name: Tempreon (or anything you'd like)
URL: https://api.tempreon.com/functions/v1/tempreon-mcp/mcp

The /mcp suffix at the end is required. It's the MCP endpoint path defined by the protocol. Leaving it off is the single most common connection error.

3. Select the transport type

If your client asks, choose Streamable HTTP (sometimes labeled "HTTP," "Remote HTTP," or "SSE"). Do not choose stdio — that transport is for local MCP servers running as subprocesses on your machine. Tempreon is hosted.

4. Select the authentication type

If your client prompts you, choose OAuth (sometimes labeled "OAuth 2.0," "OAuth 2.1," or "Automatic"). You should not need to paste any token, API key, or client ID. Tempreon supports dynamic client registration, which means your client can authenticate itself without manual configuration.

5. Authorize the connection

Save the server. Your client will trigger an OAuth flow — a browser tab opens to Tempreon's consent screen. Sign in to your Tempreon account if you're not already, review what's being requested, and approve. You'll be redirected back to your client.

6. Confirm Tempreon is active

Your client's tool list should now include Tempreon's tools. The exact way tools appear varies — some clients show them inline in a chat, others in a sidebar, others under a slash command. Start a new session and ask something like "What Tempreon tools do you have access to?" If it answers with a list, the Bridge is live.

Connection details reference

If your client wants structured details, here's everything in one place:

Field	Value
Server URL	`https://api.tempreon.com/functions/v1/tempreon-mcp/mcp`
Transport	Streamable HTTP
Authentication	OAuth 2.1 with PKCE
Client registration	Dynamic (RFC 7591) — no pre-registration required
Server type	Remote / hosted

For clients that accept a JSON configuration:

{
  "mcpServers": {
    "tempreon": {
      "url": "https://api.tempreon.com/functions/v1/tempreon-mcp/mcp"
    }
  }
}

When your client opens the authorization flow, you'll land on a Tempreon-branded consent screen. It tells you:

Which client is asking for access (e.g., "Claude Desktop," "Cursor," or whatever your client identifies itself as)
What scope of access it's requesting — Tempreon uses a single person-scoped permission; your client can only read and write data that belongs to your account
How to approve or deny — one click

After you approve, your browser redirects back to your client. The client stores an access token and uses it on every request. Your Tempreon password is never shared with the client. If you later want to revoke the connection, you can remove it from your Bridges dashboard or delete the configuration in your client.

What becomes available after connecting

Once the Bridge is active, your client has access to Tempreon's toolset — scoped to your subscription tier.

Free tier gets a curated set of essential tools for everyday use: starting and ending sessions, searching your knowledge, remembering new insights, managing tasks, and storing files. Enough to prove the value before you commit.
Standard and above unlock the full tool library: Cascades (multi-step orchestration), Decision Gates (behavioral rules), People Intelligence, and direct access to the lower-level building blocks for power workflows.

Whichever tier you're on, session_start is the best first call. It loads your Core Imprint, active preferences, and recent work. From there, your client picks the right tools for what you're doing.

Your data is scoped to your account through the OAuth token. No one else can reach it through your Bridge, and you can't reach anyone else's.

Universal troubleshooting

These five issues cover almost every connection problem, regardless of which client you're using.

Connection fails immediately. The URL is the usual culprit. Confirm it ends in /mcp. The full URL is https://api.tempreon.com/functions/v1/tempreon-mcp/mcp — no trailing slash, no query parameters.

Browser window for authorization doesn't open. Your client has to hand off to your default browser for the OAuth step. Check that pop-up blockers aren't blocking the authorization window, and that you have a working default browser set at the OS level. Some CLI clients print the authorization URL in the terminal instead of opening it — look for an https:// link in the output and paste it into your browser manually.

Tools aren't appearing in the client. Restart your client session after connecting — some clients only refresh their tool list at session start. If you're on the free tier, you'll see the curated essential toolset rather than the full library; that's expected.

"Unauthorized" or "token expired" errors. Disconnect the Bridge in your client and reconnect it. The OAuth flow will run again and issue fresh tokens.

Client says MCP isn't supported, or only supports local (stdio) servers. Check your client's release notes for "remote MCP," "HTTP MCP," or OAuth support. Some clients can bridge the gap with a shim like mcp-remote (which runs locally and proxies to a remote server), but that's a workaround. Native remote MCP support is the supported path.

For client-specific troubleshooting, see Troubleshooting Connections.

MCP-compatible clients

The Model Context Protocol was announced by Anthropic on November 25, 2024 as an open standard for connecting AI assistants to external context and tools. Adoption has been rapid — most major AI clients now support remote MCP servers over OAuth.

If a client speaks MCP with OAuth, Tempreon works with it. The table below lists clients that publicly support MCP. Entries we've fully tested end-to-end are linked to their setup guide; the rest should work using the universal flow above.

Hosted AI assistants

Client	MCP Support	Status with Tempreon
Claude Desktop	Yes (native, from 2024 launch)	Fully tested — setup guide
Claude Code	Yes (native)	Fully tested — setup guide
Claude.ai web	Yes (Connectors, shared with Desktop)	Fully tested — connect once via Desktop, appears across surfaces
Codex (OpenAI)	Yes (Mac app, IDE extension, CLI — shared `config.toml`)	Fully tested — setup guide
ChatGPT (Plus, Pro, Business, Enterprise, Education)	Yes (since Sept 9, 2025)	Fully tested — setup guide — requires enabling Developer Mode

Claude's ecosystem behavior: Once you add Tempreon as a connector in Claude, it shows up across Claude Desktop, claude.ai, and the Claude mobile app automatically. Claude Code reads its own .mcp.json, so that one needs a separate setup.

ChatGPT's Developer Mode requirement: ChatGPT's MCP support is powerful but gated behind Settings → Connectors → Advanced → Developer Mode. That extra step is a current quirk of OpenAI's rollout, not a Tempreon limitation.

Code editors and IDEs

Client	MCP Support	Status with Tempreon
Cursor	Yes (native, early 2025)	Setup guide — documented, community-verified, not yet fully tested by Tempreon
Windsurf	Yes (native)	Compatible via universal flow
Zed Editor	Yes (native)	Compatible via universal flow
VS Code (GitHub Copilot agent mode)	Yes	Compatible via universal flow
JetBrains IDEs (IntelliJ, PyCharm, WebStorm, etc.)	Yes (integrated MCP server, from 2025.2)	Compatible via universal flow
Cline	Yes (native)	Compatible via universal flow
Continue.dev	Yes (native)	Compatible via universal flow
Replit	Yes	Compatible via universal flow

Agent frameworks and SDKs

Framework	MCP Support	Status with Tempreon
OpenAI Agents SDK	Yes (since early 2025)	Apps built on the SDK can connect to Tempreon as an MCP server
Anthropic Agent SDK	Yes (native)	Tempreon is designed for this integration path
Mastra, LangChain, LlamaIndex, CrewAI	Varying MCP client support	Check your framework's current MCP docs

Frontier models and emerging clients

As more frontier models (Gemini, Kimi, DeepSeek, Llama-based runtimes, and others) add MCP client capabilities, the connection pattern stays the same: point it at https://api.tempreon.com/functions/v1/tempreon-mcp/mcp, pick OAuth, authorize. If your favorite model or client isn't listed above, check its documentation for "MCP client" or "remote MCP" support — and let us know so we can add it to this catalog.

When to use a client-specific guide

For clients we've tested end-to-end, the specific guide shows you exact menu paths, screenshots, and any client-specific quirks:

Claude Desktop & Web — Connectors flow
Claude Code — .mcp.json configuration
Codex (ChatGPT) — ~/.codex/config.toml configuration
ChatGPT — Developer Mode and custom app flow
Cursor — Settings → MCP

Once you've connected a client we haven't documented, let us know — we add guides as we see adoption.

Try saying...

Once Tempreon is connected, your AI knows about the Bridge. Try:

"What Tempreon tools do you have access to?" — Confirms the connection and shows what's available on your tier
"Start my session." — Triggers session_start to load your Core Imprint and active context
"Search my knowledge base for [topic]." — Exercises the knowledge retrieval path end-to-end