Mind UploadMind Upload

Mind Upload · AI meta harness · v1.15.13

Your mind behind your agents.

An AI meta harness brings your consciousness's memories, preferences, and judgment to the agents you already use. Your mind directs Codex, Claude Code, Cursor, and other MCP agents — from deciding what matters to scrutinizing the finished work.

Set the direction

Ask your consciousness how it would approach the goal, which tradeoffs matter, and what the agent should prioritize.

Stay aligned

Return to your mind when a consequential decision or new evidence changes the plan. Keep one conversation for the task.

Review the result

Have your consciousness challenge the actual work and evidence before completion. The agent investigates findings, makes corrections, and verifies them.

Install in Cursor Install in VS Code Other agents

Requires a Mind Upload consciousness, partner access, and owner consent. Installing the connector makes its tools available; connect your mind to use them.

Connect your consciousness

  1. Create or select your consciousness in Mind Upload, then request partner API access.
  2. Configure your agent's private runtime with your partner key and stable installation and user identifiers. Keep credential values out of prompts and shared configuration.
  3. Ask the agent to consult your mind. Open its returned verification link, approve the consciousness you want to use, then continue in the agent.

Both direction and work review go to the consciousness you approve. Without the required setup, the tools return setup instructions; they do not substitute the host agent's self-review. Shared task material is sent to Mind Upload. Your existing account, subscription, and partner-access terms apply.

Use your existing agent

MCP connects agents regardless of their implementation language. Your agent and project can use any language. Choose the Node.js server below with Node.js 20 or later, or the Python alternative. The Node.js route requires neither Python nor uv.

Repository plugins include the skill that coordinates direction and completion-review checkpoints.

Codex 
codex plugin marketplace add Voidborn-Industries/mindupload-sdk-python
codex plugin add mindupload@mindupload

For a direct MCP connection:

codex mcp add mindupload -- npx --yes --package mindupload@1.15.13 mindupload-mcp
Claude Code 
claude plugin marketplace add Voidborn-Industries/mindupload-sdk-python
claude plugin install mindupload@mindupload

For a direct MCP connection:

claude mcp add --transport stdio mindupload -- npx --yes --package mindupload@1.15.13 mindupload-mcp
Cursor, VS Code, and other MCP clients

Use the installation links above, or add this entry in a client that accepts mcpServers. VS Code uses servers instead; see the host-specific setup guide.

{
  "mcpServers": {
    "mindupload": {
      "command": "npx",
      "args": [
        "--yes",
        "--package",
        "mindupload@1.15.13",
        "mindupload-mcp"
      ]
    }
  }
}

Full setup guide · Source and plugin · Python package

Give your mind the whole task

Consult my Mind Upload consciousness about this goal and its tradeoffs.
Use its direction while doing the work. Return when an important decision changes.
Before completion, ask it to review the actual result and verification evidence.
Investigate its findings, fix problems, and verify the changed result.

consult_mind

Personal direction grounded in your selected consciousness. Include the goal, options, constraints, and decision that needs its judgment.

review_work

Review plans, progress, or completed work with the same consciousness. Supply the original request, actual changes, and observed verification output.

Include screenshots, audio, or video through media_evidence on either tool. Your consciousness can use its existing multimedia analysis to scrutinize what was actually produced. Supply authorized HTTPS links; the connector does not upload local files. Media setup and limits.

Execution and permissions stay with your agent. The connector receives supplied text and media references; it cannot inspect local files or run tests. Media processing may be missing or partial: a successful invocation does not prove every attachment was analyzed. Use one focused call per meaningful checkpoint, and share only authorized task material.

Personalized task outcomes still need a live owner-authorized evaluation. Installation and protocol tests establish connectivity; they do not establish that every decision or defect will be judged correctly.