使用说明
Galan
Quantum computing for your AI assistant.
Galan is an MCP server that gives Claude access to quantum computing — circuit analysis, paper search, QUBO solving, hardware pricing, and circuit execution on real QPUs.
No SDK to learn. No API docs to read. Just connect and ask.
Connect
Claude.ai / Claude Desktop (Remote)
Go to claude.ai/settings/connectors and add:
https://garm.valiant-quantum.com/mcp
Select Streamable HTTP as transport type. Six tools become available immediately.
Claude Desktop (Local Client)
Alternatively, install the local client:
pip install galan-mcp
galan init
This writes the config to claude_desktop_config.json automatically. Restart Claude Desktop.
Tools
compile_circuit
Analyze a quantum circuit. Accepts QASM3, Qiskit, PennyLane, or Cirq code.
Returns: circuit profile (qubits, gates, depth, algorithm class), hardness prediction via the N(N−1)/2 pair-counting bound, optimization suggestions, and backend compatibility with estimated fidelity.
Example prompt: "Analyze this QAOA circuit and tell me which hardware can run it."
research
Search 7,000+ quantum computing papers from arXiv, Qiskit, PennyLane, Classiq, D-Wave, and peer-reviewed publications.
Returns: matched papers with confidence scores, algorithm classifications, key findings, and framework references.
Example prompt: "Find papers on VQE ansatz optimization for molecular simulation."
solve_qubo
Solve a QUBO (Quadratic Unconstrained Binary Optimization) problem with hardness analysis and solver comparison.
Returns: best solution, objective value, solver comparison (simulated annealing vs greedy), hardness prediction, and QPU recommendation.
Example prompt: "Solve this 10×10 QUBO matrix and tell me if a quantum computer would help."
get_pricing
Compare quantum hardware pricing across all providers. Optionally pass a circuit to get a concrete cost comparison.
Returns: per-provider pricing models, Aleta index, merit-order ranking, and circuit-specific cost estimates.
Example prompt: "How much would it cost to run a 14-qubit VQE circuit on different backends?"
list_backends
List all available quantum backends with specifications.
Returns: backend IDs, technology type, circuit budget, gate fidelity, connectivity info.
Example prompt: "Which quantum computers are available right now?"
run_circuit
Run a quantum circuit on a QPU or simulator. Accepts QASM3, Qiskit, PennyLane, or Cirq code.
Returns: measurement counts, probabilities, statistics (entropy, dominant states), execution metadata.
Example prompt: "Run this circuit on the simulator with 4096 shots."
Hardness Prediction
Galan includes the N(N−1)/2 gate-count bound for circuit hardness estimation:
For any Hamiltonian with kinetic energy p² in N-qubit binary encoding, the free propagator requires exactly N(N−1)/2 entangling gates. This structural bound predicts circuit feasibility on current hardware before compilation.
Reference: Hinderink, D. (2026). Pair-Counting Scaling of the Quantum Chaos Threshold. Valiant Quantum.
Available Backends
| Backend | Technology | Qubits | 2Q Fidelity | Connectivity |
|---|---|---|---|---|
| Quantinuum H2 | Trapped ion | 56 | 99.9% | All-to-all |
| IonQ Forte | Trapped ion | 36 | 99.7% | All-to-all |
| IonQ Aria | Trapped ion | 25 | 99.5% | All-to-all |
| AQT IBEX Q1 | Trapped ion | 12 | 98.9% | All-to-all |
| IBM Torino | Superconducting | 133 | 99.5% | Heavy-hex |
| IBM Marrakesh | Superconducting | 156 | 99.5% | Heavy-hex |
| Rigetti Ankaa-3 | Superconducting | 84 | 99.5% | Square-octagon |
| IQM Garnet | Superconducting | 20 | 99.0% | Square lattice |
| Simulator | Classical | Unlimited | 100% | All-to-all |
Pricing
Galan normalizes pricing across providers using the Aleta Index (EUR/Aleta). Current rates:
- IBM — Free via Open Access (10 min/month QPU time)
- IQM — $0.30/task + $0.00145/shot (AWS Braket)
- IonQ — Token-based, min ~$12/execution (Azure Quantum)
- Quantinuum — HQC-based, subscription required
- Simulator — Always free
Use get_pricing with a circuit to see exact costs.
Privacy
- Circuits are anonymized before analysis (comments, variable names, string literals stripped)
- All data encrypted in transit (HTTPS/TLS 1.3)
- No circuit data is stored after execution
- Usage is logged for billing purposes only (tool name, problem size, timestamp)
Issues
Found a bug? Have a feature request? Open an issue.
About
Galan is part of the Garm quantum computing platform by Valiant Quantum. The circuit analysis engine is powered by Nathan, the quantum research optimizer from Arvak.
