ENVI/SARScape MCP Server

FastMCP server and standalone scripts for discovering a local ENVI Task Engine installation, checking SARScape task visibility, and running ENVI/SARScape InSAR DEM workflow steps.

One-Command Install

For Windows users with ENVI/SARScape, Python Launcher, and Git installed:

$script = "$env:TEMP\install-envi-mcp.ps1"
Invoke-WebRequest "https://raw.githubusercontent.com/Di0105/envi-mcp-server/main/scripts/install_user.ps1" -OutFile $script
powershell -ExecutionPolicy Bypass -File $script

This clones the repo to %USERPROFILE%\envi-mcp-server, creates .venv312, installs the MCP extra, runs connectivity, registers the VS Code user-level MCP server, and installs the workspace-agnostic envi-sarscape agent.

After install, open any VS Code workspace and choose the envi-sarscape agent.

The local acceptance target is:

1. Python can import envipyengine.
2. The server can auto-detect or use a configured ENVI 5.6 Task Engine path.
3. Engine("ENVI") starts successfully.
4. ENVI tasks can be listed.
5. SARScape/SAR-related tasks are visible.

The default Windows shortcut discovery hint is:

%USERPROFILE%\Desktop\ENVI 5.6 (64-bit).lnk

Project Layout

envi-mcp-server/
??? pyproject.toml
??? README.md
??? src/envi_mcp/
?   ??? __init__.py
?   ??? engine.py
?   ??? pipeline.py
?   ??? schemas.py
?   ??? server.py
?   ??? tools/
??? scripts/
?   ??? check_connectivity.py
?   ??? run_insar_dem.py
?   ??? config_template.yaml
?   ??? compare_dems.py
??? tests/
??? .vscode/mcp.json

Setup

For connectivity checks and standalone scripts, Python 3.9+ is enough. Use the Python environment that can import ENVI Task Engine packages. If ENVI provides its own Python, run these commands from that environment.

python -m pip install -e .[dev]

For the MCP server, use Python 3.10+ because fastmcp requires it:

py -3.12 -m venv .venv312
.\.venv312\Scripts\python.exe -m pip install -e .[mcp,dev]

Optional DEM comparison dependencies:

python -m pip install -e .[compare]

If auto-discovery misses ENVI, set ENVI_ENGINE to the full taskengine.exe path:

$env:ENVI_ENGINE = "C:\????\taskengine.exe"
$env:PYTHONUTF8 = "1"

Discovery checks, in order:

• ENVI_ENGINE
• explicit script/manager path
• common Harris, NV5, Exelis, and ITT install paths
• Windows registry uninstall entries
• the ENVI 5.6 desktop shortcut
• PATH

Connectivity Check

Run the real local acceptance script:

python scripts/check_connectivity.py --json connectivity_report.json
.\.venv312\Scripts\python.exe scripts/check_connectivity.py --json connectivity_report_py312.json

Success requires envipyengine import, ENVI startup, a non-empty task list, and at least one SARScape/SAR-related task.

For a quick task-name inspection after the first successful connection, use the MCP tools list_envi_tasks("SAR") and list_envi_tasks("SARscape").

MCP Configuration

This repo includes a project-level config at .vscode/mcp.json:

{
  "servers": {
    "envi-sarscape": {
      "type": "stdio",
      "command": "${workspaceFolder}\\.venv312\\Scripts\\python.exe",
      "args": ["-m", "envi_mcp.server"],
      "cwd": "${workspaceFolder}",
      "env": {
        "PYTHONPATH": "${workspaceFolder}/src",
        "PYTHONUTF8": "1"
      }
    }
  }
}

Make sure the python resolved by this MCP config is a Python 3.10+ environment with python -m pip install -e .[mcp] already run.

To register this server globally for all VS Code workspaces and install the user-level agent:

.\.venv312\Scripts\python.exe scripts/register_vscode_agent.py --repo-root . --python .\.venv312\Scripts\python.exe

The global agent template lives at agents/envi-sarscape.agent.md and is rendered into %APPDATA%\Code\User\prompts\envi-sarscape.agent.md.

User-level template:

{
  "servers": {
    "envi-sarscape": {
      "command": "python",
      "args": ["-m", "envi_mcp.server"],
      "cwd": "C:\\Users\\<you>\\envi-mcp-server",
      "env": {
        "PYTHONPATH": "C:\\Users\\<you>\\envi-mcp-server\\src",
        "PYTHONUTF8": "1",
        "ENVI_ENGINE": "C:\\????\\taskengine.exe"
      }
    }
  }
}

ENVI_ENGINE is optional when auto-discovery works.

MCP Tools

The server registers 37 tools across five modules:

• General: connectivity, task discovery, task listing, task info, exact task execution.
• SAR import: Sentinel-1, RADARSAT-2, ALOS/PALSAR, TerraSAR-X, COSMO-SkyMed, generic SAR import.
• Interferometry: baseline, coregistration, interferogram, coherence, adaptive filtering, unwrapping, GCP, refinement/reflattening, phase-to-height, geocoding, mosaic, flat-earth removal, multilooking.
• Postprocess: terrain correction, void fill, smoothing, GeoTIFF export, quality mask, statistics, DEM comparison, raster clip.
• Utils: pipeline step listing, input path validation, output workspace creation, pipeline state loading, full InSAR DEM pipeline.

SARScape task names differ across installs, so logical tools search installed ENVI task names by exact candidates and fallback keywords. If a wrapper cannot find a matching task, run list_envi_tasks to confirm the local name, then call run_envi_task directly.

InSAR DEM Script

Copy and edit the config template:

python scripts/run_insar_dem.py --config scripts/config_template.yaml --dry-run
python scripts/run_insar_dem.py --config scripts/config_template.yaml --resume
python scripts/run_insar_dem.py --config scripts/config_template.yaml --skip-to coherence
python scripts/run_insar_dem.py --config scripts/config_template.yaml --only unwrap

The script writes:

• state.json
• pipeline.log
• pipeline_result.json

Pipeline order is fixed as:

baseline -> coregistration -> interferogram -> coherence -> filter -> unwrap -> gcp -> refine -> phase_to_height -> geocode -> mosaic

coherence is always produced before unwrap, and resumed runs require the needed upstream output keys to exist in state.json.

DEM Comparison

After installing the comparison extra, run:

python scripts/compare_dems.py --dem C:\data\dem.tif --reference C:\data\reference_dem.tif --output-dir C:\data\dem_compare

Outputs:

• accuracy_report.txt
• diff_map.png
• diff_histogram.png
• scatter.png

Tests

Mock tests do not require a real ENVI license:

python -m pytest

They cover engine discovery priority, task filtering, task metadata serialization, task success/failure handling, pipeline state load/save, and skip/resume/only logic.

Notes

• ENVI/SARScape may require an available license; Engine("ENVI") can consume one.
• Actual InSAR DEM correctness needs a real SLC pair and reference DEM.
• SARScape task names are version-dependent; inspect local names after connectivity succeeds.
• Keep project and intermediate output paths ASCII-only when using C/C++ geospatial libraries on Windows.