Claude Desktop and a Relay: Everyday Drafting

§1 Start with the right connection model

Claude Desktop can be a convenient workspace for drafting emails, project updates, and meeting follow-ups. Adding a relay gives that workspace access to an external generation service, but the connection needs one important clarification.

Claude Desktop does not provide a general setting that replaces its built-in assistant with an arbitrary API endpoint. Instead, use a local Model Context Protocol, or MCP, bridge that exposes the relay as a tool. The desktop assistant can call that tool and use its output while helping you revise.

Fluxio fits this setup when you want pay-per-token access through an OpenAI-compatible API. Its streaming-first relay lets a compatible client receive output token by token. Your desktop subscription and relay usage remain separate.

§2 Step 1: Prepare your relay credentials

Create a Fluxio API key and store it securely. The API base is https://api.fluxioapi.com; for an OpenAI-style client, use https://api.fluxioapi.com/v1 as the base URL.

• – Keep the key out of prompts, shared documents, screenshots, and source control.
• – Check available models, current prices, and account limits before testing.
• – Choose one available model for your first drafting workflow, such as Claude Haiku 4.5, if it appears in the relay catalog.

Use the exact model identifier supplied by the provider. A display name is not necessarily the string your bridge must send in an API request.

§3 Step 2: Choose a compatible MCP bridge

Select a trusted local MCP server that explicitly supports calling an OpenAI-compatible chat endpoint. It should let you configure a base URL, API key, and model identifier, and expose a text-generation tool to Claude Desktop.

Check its documentation for local standard-input/output transport, installation requirements, and credential handling. An ordinary API client is not automatically an MCP server, and an arbitrary MCP server will not necessarily support relay requests.

For everyday drafting, prefer a narrow tool that accepts writing instructions and returns generated text. Avoid granting filesystem or shell access unless your workflow genuinely requires it.

§4 Step 3: Register the bridge in Claude Desktop

Install the bridge using its documented procedure. In Claude Desktop, open the developer settings for local MCP servers and edit the configuration through the available configuration control. Labels and locations can vary by release.

Add the launch command, arguments, and environment variables specified by the bridge documentation. Configure its OpenAI-compatible base URL as https://api.fluxioapi.com/v1, supply your API key through its supported secure mechanism, and set the selected model identifier.

Save the configuration and restart Claude Desktop if required. Confirm that the server connects and its generation tool appears. If it fails, inspect the local server logs without copying credentials into a support conversation.

§5 Step 4: Run a small drafting test

Start with fictional, non-sensitive material. Ask: “Use the relay drafting tool to write a friendly, 120-word project update. Include completed work, one blocker, and next steps. Do not invent dates or commitments.”

Approve the tool call if Claude Desktop requests permission. Verify that the relay receives the request and that the returned draft appears in the conversation.

Streaming at the API does not guarantee live token-by-token display inside Claude Desktop. The bridge and desktop tool interface determine whether you see incremental output or a completed tool result.

§6 Step 5: Turn the test into a daily routine

Use a repeatable brief: audience, purpose, source facts, tone, length, and details that must remain unchanged. Explicitly request the relay tool when you want external generation rather than an ordinary desktop response.

• – For email, provide the recipient context and the action you want them to take.
• – For status updates, separate confirmed facts from unresolved questions.
• – For revisions, ask for one targeted change instead of regenerating everything.

Review names, numbers, promises, and confidential details before sending. Tool inputs go to the relay and its upstream processing path, so follow your organization's data policies. Keep requests focused to control token costs, and check usage after your first few sessions.

Sign up for Fluxio to try your first relay-backed draft.