vCFD

Voice-driven CFD: Grok Voice + MCP + OpenFOAM + ParaView.

Speak natural language commands; a Grok voice agent calls vCFD MCP tools that set up and run OpenFOAM cases, then summarize results and export data for ParaView on your Mac.

JARVIS for CFD (pragmatic edition): full-duplex speech → tool calls → async solvers → spoken engineering summaries. Not movie magic—solid async job handling and careful tool design.

Architecture

Human (speech)
    ↕ Speech-to-Speech
Grok Voice / Voice Agent Builder
    ↓ Remote MCP (HTTP/SSE + auth)
vCFD MCP Server (Python / FastMCP)
    ↓ subprocess or Docker
OpenFOAM cases & solvers
    ↓ foamToVTK / case zip
ParaView (Mac)

vCFD does not modify OpenFOAM. It launches OpenFOAM executables and edits case dictionaries. OpenFOAM remains a separate GPLv3 runtime (see NOTICE).

Related MCP server: ANSYS MCP Server

Quick start (Mac Silicon)

Requirements

• Python 3.12+ and uv

• Optional: Docker (for real OpenFOAM; mock mode works offline)

• Optional: ParaView (visualization)

• Grok Build for agentic development / local MCP

Install

git clone https://github.com/gpu7/vCFD.git
cd vCFD
uv sync

Run tests (mock OpenFOAM)

uv run pytest -q

Run MCP server (stdio)

export VCFD_OPENFOAM_MODE=mock   # or docker | native
./scripts/dev_stdio.sh

Attach to Grok Build

grok mcp add vcfd -- uv run --directory /path/to/vCFD vcfd-mcp

Or configure in ~/.grok/config.toml:

[mcp_servers.vcfd]
command = "uv"
args = ["run", "--directory", "/path/to/vCFD", "vcfd-mcp"]
env = { VCFD_OPENFOAM_MODE = "mock" }
enabled = true

First case: Hagen–Poiseuille pipe

Through MCP tools (or voice):

1. create_case(name="pipe1", template="hagen_poiseuille")

2. set_pipe_parameters(case_name="pipe1", radius=0.05, bulk_velocity=0.1)

3. generate_mesh(case_name="pipe1") → check_mesh(...)

4. start_solver(case_name="pipe1") → get_job_status(job_id=...)

5. summarize_results / compare_to_analytical

6. export_for_paraview → open zip/VTK in ParaView

OpenFOAM modes

See .env.example for all settings.

Voice agent (Grok Voice Agent Builder)

1. Deploy public MCP (EC2: ./deploy/aws/launch-ec2.sh).

2. Follow prompts/voice_agent_builder.md — console steps + MCP fields.

3. System prompt: prompts/voice_agent_system.md.

4. Code / paste helpers: voice/session.example.json, scripts/voice_session_snippet.py --from-latest.

./deploy/aws/launch-ec2.sh
uv run python scripts/voice_session_snippet.py --from-latest
# Paste URL + Bearer token into Voice Agent Builder remote MCP

AWS EC2

Automated launch scripts (AWS CLI on your Mac):

./deploy/aws/launch-ec2.sh --dry-run   # plan only
./deploy/aws/launch-ec2.sh             # launch Ubuntu + Docker + Caddy + MCP
./deploy/aws/status.sh
./deploy/aws/teardown.sh

Details: deploy/aws/README.md.

Project layout

src/vcfd/           # MCP server, OpenFOAM runner, analytics
cases/templates/    # OpenFOAM case templates (not OF source)
tests/              # pytest
deploy/             # Docker, systemd, AWS notes
prompts/            # Voice agent system prompt
scripts/            # dev helpers

License

• vCFD code: MIT

• OpenFOAM: GPLv3 (external dependency — not distributed as part of this repo’s source)

You may charge for applications or services built with vCFD as an external controller of OpenFOAM. Do not link or redistribute modified OpenFOAM without complying with GPLv3.

Status

• Mock OpenFOAM pipeline: tested (uv run pytest)

• Real OpenFOAM via Docker (opencfd/openfoam-default:2506, arm64-capable): mesh + simpleFoam + ParaView export verified on Mac Silicon

• Grok Build project MCP: .grok/config.toml (grok mcp list)

• Next: Voice Agent Builder / remote HTTP MCP on EC2, richer field sampling, more templates

Docker end-to-end

# Requires Docker Desktop running and image pull once:
docker pull opencfd/openfoam-default:2506
./scripts/e2e_docker.sh

Results land under cases/workspace/ (gitignored). Open the .foam or VTK/ export in ParaView.

Grok Build MCP

Project config is already set (docker mode):

grok mcp list
# tools available after restart / new session in this directory