본문으로 건너뛰기

Environment Setup

FluxLoop automatically detects and uses your project’s Python environment so experiments, CLI commands, and the Integration Assistant run inside the correct interpreter.

Python Requirements
• FluxLoop CLI → Python 3.8+
• FluxLoop SDK & MCP → Python 3.11+ (required for Flux Agent)

Quick Setup​

  1. Create / activate a venv 
    python3.11 -m venv .venv
    source .venv/bin/activate   # Windows: .venv\Scripts\activate
  2. Install FluxLoop packages 
    pip install fluxloop-cli fluxloop fluxloop-mcp
  3. Select execution mode
    • Command Palette → FluxLoop: Select Environment
    • Choose Auto (detect project environment) or Workspace only
    • If dependencies are missing the wizard offers automatic install or manual instructions
  4. Verify detection
    • Command Palette → FluxLoop: Show Environment Info
    • Confirm Python/fluxloop/fluxloop-mcp paths point to .venv
    • View → Output → FluxLoop for detailed logs
  5. Set Target Source Root…
    • Projects view → Target Source Root…
    • Select the directory containing your actual agent code (Flux Agent uses this for analysis)

FluxLoop now uses this environment for experiments, CLI commands, and Integration Assistant runs.

Automated Setup Script​

For faster onboarding run:

cd your-project
bash packages/cli/scripts/setup_fluxloop_env.sh \
  --python python3.11 \
  --target-source-root ./src

The script:

  • Creates .venv (if needed)
  • Installs CLI, SDK, MCP
  • Writes .vscode/settings.json with fluxloop.executionMode, fluxloop.targetSourceRoot, fluxloop.projectRoot
  • Updates configs/project.yaml → source_root
  • Runs fluxloop doctor

The VS Code project wizard exposes this as Run Setup Script.

Execution Modes​

Choose how FluxLoop locates executables:

ModeDescriptionUse When
Auto (default)Detect .venv, Poetry, Conda, uv; fallback to PATHGeneral use
Workspace onlyRequire a project-local venv; fail otherwiseTeams enforcing per-project envs
Global PATHAlways use global executables (pipx, Homebrew)Quick prototyping
Custom executablesManually specify Python + fluxloop-mcp pathsComplex Conda/pyenv setups

Set via:

  • Command Palette → FluxLoop: Select Environment
  • Integration view → System Status → Select Environment
  • .vscode/settings.json: 
    {
      "fluxloop.executionMode": "auto",
      "fluxloop.pythonPath": "/custom/python",
      "fluxloop.mcpCommandPath": "/custom/fluxloop-mcp"
    }

Target Source Root​

Defines where FluxLoop searches for envs and source files (also used by Flux Agent).

Projects view workflow

  1. Expand your project
  2. Click Target Source Root…
  3. Choose a folder (relative paths stored automatically)
  4. FluxLoop refreshes detection and logs the results

Manual configuration

# configs/project.yaml
source_root: "./backend"

Monitoring Environment Status​

Show Environment Info​

FluxLoop: Show Environment Info → quick modal with:

  • Source root
  • Execution mode
  • Python/fluxloop/fluxloop-mcp paths
  • Detection notes (missing packages, etc.)

Output Channel​

View → Output → select FluxLoop. Every refresh prints:

[FluxLoop Env] Source root: /Users/me/src/app
[FluxLoop Env] Environment: venv
[FluxLoop Env] Python: .../.venv/bin/python
[FluxLoop Env] fluxloop: .../.venv/bin/fluxloop
[FluxLoop Env] fluxloop-mcp: .../.venv/bin/fluxloop-mcp
[FluxLoop Env] Notes:
[FluxLoop Env]  • fluxloop-mcp not found; install fluxloop-mcp in this environment.

Doctor​

Run comprehensive diagnostics:

  • Command Palette → FluxLoop: Run Doctor
  • Integration view → System Status → Run Doctor
  • Terminal → fluxloop doctor [--json]

Doctor checks Python, venv, CLI, SDK, MCP, knowledge index, and config structure.

Common Scenarios​

uv​

  1. uv venv && source .venv/bin/activate
  2. uv pip install fluxloop-cli fluxloop fluxloop-mcp
  3. (Optional) FluxLoop: Configure Execution Wrapper → uv run

pipx​

  1. pipx install fluxloop-cli / pipx install fluxloop-mcp
  2. Install fluxloop SDK in your project venv
  3. Set execution mode to Global PATH

Conda​

  1. conda create -n agent python=3.11
  2. conda activate agent
  3. pip install fluxloop-cli fluxloop fluxloop-mcp
  4. If VS Code doesn’t detect it automatically, use Custom executables and point to /opt/conda/envs/agent/bin/python

Multiple Projects​

  • Each project stores its own execution mode, target source root, wrapper, etc.
  • Switching projects triggers a fresh detection using the cached settings.

Troubleshooting​

“Environment not detected”​

  1. Run FluxLoop: Show Environment Info
  2. Check FluxLoop Output for errors
  3. Run FluxLoop: Run Doctor
  4. Ensure the selected interpreter actually contains fluxloop and fluxloop-mcp: 
    source .venv/bin/activate
    which fluxloop
    which fluxloop-mcp

“Commands use global Python instead of venv”​

  1. Ensure .venv exists under your target source root
  2. Set execution mode to Auto or Workspace only
  3. Confirm fluxloop.targetSourceRoot points at the folder containing .venv

“Flux Agent says MCP missing”​

  1. Verify MCP installed in active env
  2. Check doctor output for index path
  3. Rebuild index: packages/mcp/scripts/rebuild_index.sh
  4. Confirm fluxloop.mcpCommandPath isn’t pointing to an old installation

“Wrong project analyzed”​

  • Run Target Source Root… again and choose the correct folder
  • Ensure configs/project.yaml → source_root matches your selection
  • Integration view displays the currently active source root in System Status

Next Steps​