← All agents

Claude Code

claude mcp add --transport http qedra https://your-domain/api/mcp

Cursor ~/.cursor/mcp.json

{
  "mcpServers": {
    "qedra": {
      "url": "https://your-domain/api/mcp"
    }
  }
}

VS Code .vscode/mcp.json

{
  "servers": {
    "qedra": {
      "type": "http",
      "url": "https://your-domain/api/mcp"
    }
  }
}

Tools

Quantum lab
plan_experiment

Turns a plain-language request into a validated experiment plan (template, parameters, shots).

run_experiment

Runs Bell, GHZ, teleportation, superdense coding, Grover, Deutsch–Jozsa, Bernstein–Vazirani, Max-Cut/QAOA, random numbers or a custom circuit on the simulator; returns counts, exact probabilities and an analysis.

get_job

Reads back a job created through the MCP endpoint.

verify_receipt

Looks up on-chain receipts for a job or result commitment.

get_limits

Lists the templates and the simulator's limits.

On-chain agents (Robinhood Chain)
token_safety_report(token_address)

Investigates an ERC-20: owner powers (mint, blacklist, pause, fees, upgrades), real burns vs new mints, holder concentration, pools and the deployer. Returns verified findings with evidence plus a plain-language summary. Takes 10–120 s.

quantum_exposure(address)

Checks whether an address has revealed its public key on-chain, with a plain explanation. Contracts are checked through their owner or multisig signers. Read-only.

What agents can't do

  • Submit real hardware jobs — the IBM backend is only available in the workspace, behind human approval.
  • Sign transactions, record receipts or move funds — everything on-chain is signed by a person's own wallet.
  • Change limits or run arbitrary code — circuits use a validated JSON schema; nothing uploaded is executed.

Results are reported as measured, never smoothed. Token reports are signals, not verdicts. Jobs and reports created over MCP share one anonymous owner, so don't put anything private in them.