Detailed Analysis
Custom icon support for Claude connectors is currently not implemented, confirming the frustration expressed by developers attempting to brand their MCP (Model Context Protocol) integrations within Claude's interfaces. When users set up a custom connector via Claude Settings, the process only prompts for an MCP server URL and optional OAuth configuration — no icon upload or icon field is available at any point in the workflow. The fact that a developer's icon renders correctly in their own test dashboard but fails to appear in Claude confirms the issue is not a misconfiguration on the developer's side; it is simply a missing feature on Anthropic's end.
The technical gap is well-documented in open GitHub issues. Issue #1040 in the Model Context Protocol repository identifies that the `ServerInfo` struct returned during an MCP `initialize()` response lacks an `icon` field entirely, meaning Claude Desktop has no mechanism to read or render such data even if a developer includes it. A separate issue (#152) in the `anthropics/claude-ai-mcp` repository specifically requests that Claude.ai parse and display an `icons` array from the `serverInfo` in the `initialize` response — this request remains unimplemented as of current documentation. The YouTube video the original poster references likely predates these limitations becoming widely understood, or may depict a use case where branding appears through a different mechanism such as an OAuth consent screen rather than the connector list itself.
The absence of this feature matters disproportionately to developers building production-quality integrations for businesses or end users. Custom connectors are positioned as a way to extend Claude's capabilities into proprietary workflows and applications, and visual identity — a recognizable icon — is a basic expectation for any third-party integration in a modern software ecosystem. Without it, all custom connectors appear identically in Claude's UI, creating navigation friction and undermining the polished experience developers are trying to deliver to their own users. Free-tier users are additionally limited to a single custom connector, while Team and Enterprise plans require admin approval, meaning the icon gap affects a relatively sophisticated user base with higher expectations.
Zooming out, this limitation reflects a broader pattern in the rapid rollout of MCP as a standard. Anthropic has moved quickly to establish MCP as the dominant protocol for AI tool integration, but the ecosystem is still maturing — core quality-of-life features like icon rendering, rich metadata display, and granular permission controls are in various stages of proposal and implementation. The open GitHub issues serve as the primary signal that Anthropic is aware of these gaps and is fielding community input, though no timeline for resolution has been publicly communicated. Developers building on MCP today are effectively working with a v1 infrastructure that prioritizes functional connectivity over presentation-layer polish, a trade-off that is common in early platform ecosystems but increasingly noticeable as adoption scales.
Read original article →