Proposal for Adding Relay OCR Recognition to DeepSeek and Similar Non‑Visual Models

[sh-u-x-in]

Problem (one or two sentences)

I noticed that DeepSeek does not have a relay OCR feature, which leads to poor compatibility for plain text (where formatting and positional structure are not important), even though such text could be easily converted by a simple OCR.

Context (who is affected and when)

Users of the DeepSeek web version, or those accustomed to using visual models, may find it slightly inconvenient when switching to ZooCode.

Desired behavior (conceptual, not technical)

In the settings, allow enabling a relay MCP‑OCR. This would let the system automatically relay images to an MCP‑OCR service before sending them—the image is first sent to the MCP‑OCR, the recognised text is injected into the context, and the temporary image is also exposed so that the AI can call other, more detailed MCP‑OCR services. This avoids re‑implementing OCR; instead, the automatic relay interface is exposed to users, allowing them to plug in any MCP they trust, with the user taking responsibility for confirming the security of that MCP.

Constraints / preferences (optional)

No response

Request checklist

• This describes a specific problem with clear context and impact
• I've searched existing Issues and Discussions for duplicates

Zoo Code Task Links (optional)

No response

Acceptance criteria (optional)

No response

Proposed approach (optional)

No response

Trade-offs / risks (optional)

No response