{
  "markdown": "![PyPI - Version](https://img.shields.io/pypi/v/arcgis-mcp-bridge?color=blue)\n![PyPI - Downloads](https://img.shields.io/pypi/dm/arcgis-mcp-bridge?color=maroon)\n![CI](https://github.com/muend/arcgis-mcp-bridge/actions/workflows/ci.yml/badge.svg)\n![Python 3.11+](https://img.shields.io/badge/python-3.11%2B-blue)\n![License](https://img.shields.io/badge/license-Apache--2.0-green)\n![Tools](https://img.shields.io/badge/tools-100-orange)\n![Ruff](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/ruff/main/assets/badge/v2.json)\n![uv](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/uv/main/assets/badge/v0.json)\n[![arcgis-mcp-bridge MCP server](https://glama.ai/mcp/servers/muend/arcgis-mcp-bridge/badges/score.svg)](https://glama.ai/mcp/servers/muend/arcgis-mcp-bridge)\n[![smithery badge](https://smithery.ai/badge/muend/arcgis-mcp-bridge)](https://smithery.ai/servers/muend/arcgis-mcp-bridge)\n\n# arcgis-mcp-bridge\n\n## Quick Start\n\n`arcgis-mcp-bridge` requires Windows, a licensed ArcGIS Pro installation, and\nPython 3.11 or newer for the bridge package.\n\nInstall the bootstrap package using **one** package manager:\n\n```powershell\n# Option A — pip\npy -m pip install --upgrade arcgis-mcp-bridge\n\n# Option B — uv\nuv pip install --upgrade arcgis-mcp-bridge\n```\n\nThen clone ArcGIS Pro's Python environment:\n\n```powershell\n# The final JSON report contains the target `python_exe` path.\npy -m arcgis_mcp.setup_env\n```\n\nIf the installed console command is available on `PATH`,\n`arcgis-mcp-setup` is equivalent to `py -m arcgis_mcp.setup_env`.\n\n> **Important for Windows systems with multiple Python installations:** the\n> setup is not complete until `arcgis-mcp-bridge` is installed into the\n> reported `arcgis-mcp-env\\python.exe`. Use that same interpreter for both the\n> MCP server `command` and `ARCPY_PYTHON_PATH`. This prevents worker failures\n> caused by packages or native extensions being loaded from another Python\n> environment.\n\nSee **05 — Installation** for the complete setup and configuration.\n\n**100 declarative geoprocessing tools. Two isolated processes. One security floor.**\n\nA secure, local-first, asynchronous MCP server exposing ArcGIS Pro's ArcPy\nengine to Claude Desktop and other MCP hosts over stdio JSON-RPC.\n\nTechnical write-up: [Building a Secure MCP Bridge for ArcGIS Pro and ArcPy](https://dev.to/muend/building-a-secure-mcp-bridge-for-arcgis-pro-and-arcpy-511g)\n\n| | |\n|---|---|\n| Catalog | 100 tools · 10 verticals |\n| Tests | 86 unit tests · 86/86 passing · arcpy mocked |\n| Real runtime evidence | [Reproducible ArcGIS Pro MCP smoke benchmark](benchmarks/) |\n| Static analysis | Ruff clean · Mypy `strict` clean |\n| Transport | JSON-RPC 2.0 over stdio |\n| License | Apache-2.0 |\n\n---\n\n## Why arcgis-mcp-bridge?\n\n| Feature | arcgis-mcp-bridge | geo2004/MCP-ArcGISPro | nicogis (C#/.NET) |\n|---|---|---|---|\n| Tools | **100** | ~15 | ~10 |\n| **Dependency Sync** | **Deterministic (`uv.lock`)** | Imperative (`requirements.txt`) | Native NuGet |\n| Transport | stdio JSON-RPC | file-based IPC | Named Pipes |\n| Security Architecture | Documented PathGuard sandbox | None specified / default host access | None specified / default host access |\n| arcpy Isolation | **Two-process architecture** | Single process execution | Add-In in-process execution |\n| CI (Offline Verification) | ✅ Supported | ❌ Not available | ❌ Not available |\n| License | Apache-2.0 | MIT | MIT |\n\n---\n\n## Highlight: Sketch → GIS Pipeline\n\nHand-drawn parcel boundary → photo → geodatabase feature class.\nORB+RANSAC image registration, HSV ink segmentation, direct GDB commit.\nNo manual digitizing required.\n\n> **Demo coming soon.** To preview the sketch-to-GIS pipeline:\n> 1. Draw a polygon on paper and photograph it.\n> 2. Ask Claude: *\"Use extract_sketch_to_gis to register this photo\n>    against my basemap and commit the result to my GDB.\"*\n> 3. The feature class appears in ArcGIS Pro — no manual digitizing.\n\n---\n\n## 00 — Example Prompts\n\nAfter `health_check` succeeds, talk to Claude naturally:\n\n```\n\"Buffer all parcels in my GDB by 50 meters and save to scratch.\"\n\"List all feature classes in C:\\GIS\\city.gdb starting with 'road_'.\"\n\"Dissolve the neighborhoods layer by district_id.\"\n\"Run kernel density on crime_points with a 500-meter search radius.\"\n\"Calculate slope and aspect from the DEM at C:\\GIS\\dem.tif.\"\n\"Find the 3 nearest facilities to each incident in my network dataset.\"\n\"Check geometry on all feature classes in my GDB and repair errors.\"\n```\n\n## 01 — Core Architecture & Philosophy\n\n```mermaid\nflowchart TD\n    A[Claude Desktop / Cursor] -->|JSON-RPC over stdio| B[Layer A · MCP Protocol Host]\n    B -->|NDJSON subprocess bridge| C[Layer B · ArcPy Worker]\n    C --> D[ArcGIS Pro / ArcPy Runtime]\n```\n\n**Layer A — Async Event-Driven Server** (`arcgis_mcp/server.py`).\nFastMCP on the bridge interpreter. Owns the stdio channel, validates every\nrequest against frozen Pydantic v2 contracts, dispatches work via\n`asyncio.create_subprocess_exec` — the event loop never blocks on a\ngeoprocessing call and never holds a thread lock. Layer A contains **zero\nmodule-level `arcpy` or `cv2` imports** (verified by grep in the audit\ngate); it cannot crash on Esri's native code because it never touches it.\n\n**Layer B — Subprocess ArcPy Isolation Worker** (`arcgis_mcp/worker.py`).\nSpawned per job on the licensed ArcGIS Pro interpreter\n(`ARCPY_PYTHON_PATH`). The only place `import arcpy` is legal; `cv2` loads\nlazily inside the one vision tool that needs it. Worker stdout is rebound\nto stderr at startup — the single sanctioned stdout write is the final\nNDJSON result frame, so native ArcObjects chatter can never corrupt the\nJSON-RPC channel. A native crash terminates the worker, not the server:\nthe parent converts a non-zero exit into a structured error frame.\n \n**Declarative registry** (`arcgis_mcp/registry.py`).\nEach tool is one `ToolSpec(name, category, description, input_model,\nworker_fn, destructive)`. One generic proxy factory materializes all 100\ncatalog MCP endpoints in Layer A; one generic `run_tool` dispatcher serves\nthem in Layer B. The catalog is exposed alongside three core endpoints:\n`health_check`, `list_layers`, and `execute_spatial_tool`. Adding catalog\ntool #101 touches two files — never the runtime loops.\n\nEvery failure crossing the process boundary is classified:\n`validation` · `security` · `license` · `geoprocessing` (with the full\n`arcpy.GetMessages()` stack) · `internal`.\n\n---\n\n## 02 — The 100-Tool Census Matrix\n\n| # | Vertical | Tools | Key capabilities |\n|---|---|---:|---|\n| 1 | `map_layer_management` | 10 | .aprx maps, layer order/visibility/symbology, camera, save |\n| 2 | `data_management` | 22 | FC/GDB lifecycle, fields, Describe, Excel/GeoJSON/CSV exchange |\n| 3 | `geometry_analysis` | 23 | Overlays, dissolve/merge, selections, joins, proximity, fishnet |\n| 4 | `coordinate_reference_projection` | 4 | WKID-driven define/project for vector + raster, CRS lookup |\n| 5 | `raster_operations` | 15 | Map algebra, zonal stats, DEM slope/aspect/hillshade, hydrology |\n| 6 | `vision_analytics` | 1 | Sketch-to-GIS: ORB+RANSAC registration → HSV ink → GDB commit |\n| 7 | `export_layout` | 9 | PDF/PNG plots, DPI control, map frames, text/legend, page size |\n| 8 | `editing_topology` | 7 | Repair/check geometry, append, dedupe, diff, topology validation |\n| 9 | `network_analysis` | 4 | Service areas, routing, OD cost matrix, closest facility |\n| 10 | `spatial_statistics` | 5 | Mean center, ellipse, kernel density, Gi* hot spots, Moran's I |\n| | **Total** | **100** | |\n\nEsri extension licenses (`Spatial`, `Network`) are managed through one shared\ncontext manager and checked back in via `finally` on normal Python exception\npaths. Worker-process isolation contains native failures to the current job,\nwhile unavailable licenses return a structured error frame instead of\nterminating the MCP server.\n\n### Destructive Mutation Safety Floor\n\nTen state-mutating tools refuse to run without an explicit\n`confirm: true` payload token. The gate fires in the dispatcher **before**\nthe 10–30 s `arcpy` import is paid, and the registry refuses to even\nregister a destructive spec whose contract lacks a `confirm` field:\n\n```text\nappend_features        calculate_field        define_projection\ndelete_dataset         delete_field           delete_identical\nextract_sketch_to_gis  near_analysis          remove_layer_from_map\nrepair_geometry\n```\n\n`calculate_field` carries an additional expression-channel floor: the\ndefault `expression_type` is **ARCADE** (Esri's sandboxed expression\nlanguage), and `PYTHON3` — which executes code inside the worker — is\nrejected at the Layer-A contract boundary unless `confirm: true` is\nexplicitly supplied. `raster_calculator` expressions are constrained to a\npure map-algebra grammar (identifiers, numbers, operators; no quotes, no\ndunder access) by a contract validator.\n\n---\n\n## 03 — Automated Quality Gate & Testing\n\nLicensed-runtime evidence is reported separately in the\n[`benchmarks/`](benchmarks/) method card. Its committed result uses a real\nArcGIS Pro worker and a dedicated scratch GDB; it is not pooled with the mocked\nunit-test count or presented as validation of all 100 geoprocessing tools.\n\n**Scope, stated plainly:** the automated gate currently consists of\n**86 unit tests** spanning the PathGuard boundary, the Pydantic contracts,\nthe generic registry path-guard and registration invariants, the worker's\nerror-boundary mapping, and `Settings` environment validation. It exercises\nthe catalog's structural contracts and every security-critical seam — it does\nnot claim multi-scenario validation of the 100 geoprocessing tools themselves,\nwhich execute against a licensed ArcGIS runtime that no CI runner has.\n\n**In-memory test architecture.** `tests/conftest.py` injects `MagicMock`\nproxies into `sys.modules[\"arcpy\"]` and `sys.modules[\"arcpy.sa\"]` (with\n`CheckExtension` answering `\"Available\"`) before any package import\nresolves. The entire suite executes in well under a second, with no ArcGIS\ninstallation, no license checkout, and no Esri runtime — locally and in CI\nidentically.\n\n**Test scopes.**\n\n- `tests/test_security.py` & `tests/test_pathguard.py` — the PathGuard boundary\n  firewall, exercised against real directories via pytest's `tmp_path` fixture:\n  valid reads/writes inside the sandbox pass; traversal (`..`-segments), UNC,\n  relative, NUL-byte, reserved-device, over-length and out-of-root paths are\n  rejected; write discipline (ArcGIS dataset-name rules, overwrite opt-in) is\n  enforced.\n- `tests/test_contracts.py` — Pydantic contract enforcement: per-tool parameter\n  specs, cross-field validators, `frozen` / `extra=\"forbid\"`, and the\n  `ok`-xor-`error` invariant on the IPC envelope.\n- `tests/test_registry.py` & `tests/test_registry_guard.py` — registry stream\n  integrity plus generic `apply_path_guard` enforcement and `register`\n  invariants — every schema must be a `ToolInput` subclass, every `path_fields`\n  entry must reference a valid role, duplicate names are rejected, and every\n  destructive spec must carry its `confirm` gate.\n- `tests/test_worker.py` — `process_frame` error-boundary mapping: every failure\n  class (validation, security, license, geoprocessing, internal) maps to its\n  distinct `WorkerError.kind`.\n- `tests/test_config.py` — `Settings.from_environment` validation: required\n  variables, directory/file checks, integer bounds, and the fail-fast on a\n  missing scratch geodatabase.\n\nThe side-effect import `import arcgis_mcp.tools` in the registry test is\nwhat populates the catalog; it is `# noqa`-pinned so no linter ever strips\nit again.\n\n**Static analysis.** Ruff enforces canonical formatting plus\n`E/W/F/I/B/RUF` at 88 columns against a `py311` floor (code must parse on\nthe oldest supported interpreter — Layer B). Turkish comments are\nfirst-class: the dotless `ı`/`İ` are registered under\n`allowed-confusables`, so prose is configured around, never rewritten.\nMypy runs `strict = true` with the Pydantic plugin across all 31 source\nfiles.\n\n```bash\nmake format          # ruff format + import sorting (mutates)\nmake lint            # ruff check, mutates nothing\nmake type-check      # mypy --strict over arcgis_mcp/\nmake security-audit  # live registry inspection: path roles + confirm gates\nmake verify-all      # lint + type-check + security-audit, one gate\npython -m pytest     # 86/86\n```\n\n---\n\n## 04 — Security Framework (PathGuard Sandbox)\n\nEvery filesystem argument in every contract declares its role —\n`\"read\"`, `\"write\"`, or `\"read_list\"` — in the model's `path_fields`\nmapping. One shared enforcement function applies those declarations in\n**both** processes: Layer A pre-checks before a worker is ever spawned;\nLayer B re-validates because it never trusts its parent.\n\nTwo boundary controls:\n\n- `validate_read(raw: str)` — fully resolves the path (symlinks, `..`,\n  relative segments collapsed *before* any comparison) and requires\n  containment inside a configured `allowed_roots` directory. Existence is\n  enforced via a **deepest-existing-prefix** resolution strategy: the\n  targeted path or its filesystem-resolvable geodatabase prefix must\n  exist. This is what makes GDB-internal datasets\n  (`…\\city.gdb\\roads`) first-class — the `.gdb` container is validated on\n  the filesystem, while the logical tail is constrained to plain dataset\n  names only arcpy can resolve.\n- `validate_write(raw: str, *, overwrite: bool)` — same resolution and\n  containment, plus ArcGIS-legal dataset naming and the overwrite\n  discipline: an existing target is never replaced unless the request\n  explicitly sets `overwrite: true`.\n\nAny escape pattern — traversal sequences, UNC shares, NUL bytes, reserved\ndevice names, out-of-root targets — raises `PathSecurityError`\nimmediately: the request is answered with a structured `security` frame\nand no subprocess is ever orchestrated for it.\n\n---\n\n## 05 — 📦 Installation\n\nChoose the onboarding path that matches your use case.\n\n### Prerequisites\n\n- Windows with a licensed ArcGIS Pro installation\n- Python 3.11 or newer for `arcgis-mcp-bridge`\n- An existing writable directory for `ARCGIS_MCP_ALLOWED_ROOTS`\n- An existing file geodatabase for `ARCGIS_MCP_SCRATCH_GDB`, unless\n  `<first allowed root>\\scratch.gdb` already exists\n\n### Path A: Pure PyPI Installation — Recommended for Windows Users\n\nThis is the simplest and most reliable setup for Claude Desktop and other MCP\nhosts on Windows. The recommended configuration uses the same\n`arcgis-mcp-env\\python.exe` for both Layer A (the MCP server) and Layer B\n(the ArcPy worker).\n\nChoose one bootstrap installation command:\n\n```powershell\n# Option A — pip\npy -m pip install --upgrade arcgis-mcp-bridge\n\n# Option B — uv\nuv pip install --upgrade arcgis-mcp-bridge\n```\n\nThen clone ArcGIS Pro's Python environment:\n\n```powershell\n# The final JSON report contains the target `python_exe` path.\npy -m arcgis_mcp.setup_env\n```\n\nIf the installed console command is available on `PATH`,\n`arcgis-mcp-setup` is equivalent to `py -m arcgis_mcp.setup_env`.\n\nCopy the `python_exe` value from the JSON report and assign it below:\n\n```powershell\n$ArcGISMcpPython = \"C:\\...\\envs\\arcgis-mcp-env\\python.exe\"\n```\n\nChoose one installation command:\n\n```powershell\n# Standard installation\n& $ArcGISMcpPython -m pip install --upgrade arcgis-mcp-bridge\n\n# OR: include the optional OpenCV-based sketch-to-GIS extension\n& $ArcGISMcpPython -m pip install --upgrade \"arcgis-mcp-bridge[vision]\"\n```\n\nDo not run both commands; the second command already installs the standard\npackage together with the `vision` extra.\n\nUse `$ArcGISMcpPython` as both the MCP server interpreter and\n`ARCPY_PYTHON_PATH`. This prevents `arcgis_mcp`, Pydantic, `pydantic-core`,\nand other native dependencies from being resolved from a different Python\ninstallation.\n\n### Path B: Git Clone & Deterministic Development — GIS Contributors\n\nThis path keeps Layer A in a hermetic development environment while running\nArcPy work in a separately cloned, licensed `arcgis-mcp-env` worker.\n\n```powershell\n# 1. Clone the repository.\ngit clone https://github.com/muend/arcgis-mcp-bridge.git\ncd arcgis-mcp-bridge\n\n# 2. Create the isolated development environment.\n#    Do not use --system-site-packages: Layer A must remain independent of arcpy.\nuv venv --python \"C:\\Program Files\\ArcGIS\\Pro\\bin\\Python\\envs\\arcgispro-py3\\python.exe\"\n\n# 3. Synchronize the committed dependency resolution.\nuv sync --locked\n```\n\nChoose one worker-provisioning command:\n\n```powershell\n# Standard worker\nuv run python -m arcgis_mcp.setup_env --install-runtime-deps --project-root .\n\n# OR: worker with the optional OpenCV-based sketch-to-GIS extension\nuv run python -m arcgis_mcp.setup_env --with-vision --project-root .\n```\n\n`--with-vision` implies runtime-dependency installation, so the two commands\nshould not be run consecutively.\n\nThe setup command is idempotent, accepts `--env-name` (default:\n`arcgis-mcp-env`) and `--dry-run`, and emits a JSON report. Set\n`ARCGIS_CONDA_EXE` if ArcGIS Pro's `conda.exe` is not available on `PATH`.\n\n### Worker Interpreter Integrity\n\nLayer B is launched as:\n\n```text\nARCPY_PYTHON_PATH -m arcgis_mcp.worker\n```\n\nThe interpreter referenced by `ARCPY_PYTHON_PATH` must be able to import the\ncomplete worker stack:\n\n```text\narcgis_mcp\npydantic\npydantic_core\narcpy\n```\n\nFor a first-time Windows installation, use the same\n`arcgis-mcp-env\\python.exe` for the server `command` and\n`ARCPY_PYTHON_PATH`. Separate server and worker environments remain supported\nfor development, but the worker interpreter must contain its own compatible\ninstallation of `arcgis-mcp-bridge` and all runtime dependencies.\n\nRun this preflight check before configuring the MCP host:\n\n```powershell\n$ArcGISMcpPython = \"C:\\...\\envs\\arcgis-mcp-env\\python.exe\"\n\n& $ArcGISMcpPython -c \"import sys, arcgis_mcp, pydantic, pydantic_core; print(sys.executable); print('Bridge runtime OK')\"\n& $ArcGISMcpPython -c \"import arcpy; print('ArcPy', arcpy.GetInstallInfo().get('Version'))\"\n```\n\n### Environment Variables\n\n| Variable | Required | Purpose |\n|---|---|---|\n| `ARCPY_PYTHON_PATH` | yes | Absolute path to the licensed worker `python.exe`; it must resolve `arcgis_mcp`, Pydantic/`pydantic_core`, and ArcPy |\n| `ARCGIS_MCP_ALLOWED_ROOTS` | no | Windows `;`-separated PathGuard boundary roots; defaults to `~/Documents/ArcGIS/Projects` |\n| `ARCGIS_MCP_SCRATCH_GDB` | no | Default output workspace; if omitted, defaults to `<first allowed root>\\scratch.gdb`; the GDB must already exist |\n| `ARCGIS_MCP_LOG_FILE` | no | Optional rotating log-file path |\n| `ARCGIS_MCP_LOG_LEVEL` | no | `DEBUG`, `INFO`, `WARNING`, or `ERROR`; default `INFO` |\n| `ARCGIS_MCP_TOOL_TIMEOUT` | no | Positive per-job timeout in seconds; default `600` |\n| `ARCGIS_MCP_MAX_WORKERS` | no | Concurrent ArcPy worker ceiling; default `2`, protecting license seats and RAM |\n\n### Claude Desktop Configuration\n\n`ARCPY_PYTHON_PATH` is required in every configuration and must point to the\nlicensed interpreter reported by the setup command.\n\nReplace every placeholder path below with an existing path on your machine.\nThe scratch geodatabase must already exist.\n\n#### Option 1: Unified PyPI Environment — Recommended on Windows\n\nUse the same interpreter for the MCP server and ArcPy worker:\n\n```json\n{\n  \"mcpServers\": {\n    \"arcgis-mcp-bridge\": {\n      \"command\": \"C:\\\\...\\\\envs\\\\arcgis-mcp-env\\\\python.exe\",\n      \"args\": [\n        \"-m\",\n        \"arcgis_mcp.server\"\n      ],\n      \"env\": {\n        \"ARCPY_PYTHON_PATH\": \"C:\\\\...\\\\envs\\\\arcgis-mcp-env\\\\python.exe\",\n        \"ARCGIS_MCP_ALLOWED_ROOTS\": \"C:\\\\GIS\\\\Data;C:\\\\Workspace\",\n        \"ARCGIS_MCP_SCRATCH_GDB\": \"C:\\\\GIS\\\\Data\\\\scratch.gdb\",\n        \"ARCGIS_MCP_MAX_WORKERS\": \"2\"\n      }\n    }\n  }\n}\n```\n\nThe `command` and `ARCPY_PYTHON_PATH` values should be identical in this\nconfiguration. Use the `python_exe` value returned by the setup command.\n\n#### Option 2: Local Git Development Environment\n\nUse the repository `.venv` for Layer A and the provisioned\n`arcgis-mcp-env` for Layer B:\n\n```json\n{\n  \"mcpServers\": {\n    \"arcgis-mcp-bridge\": {\n      \"command\": \"C:\\\\path\\\\to\\\\arcgis-mcp-bridge\\\\.venv\\\\Scripts\\\\python.exe\",\n      \"args\": [\n        \"-m\",\n        \"arcgis_mcp.server\"\n      ],\n      \"env\": {\n        \"ARCPY_PYTHON_PATH\": \"C:\\\\...\\\\envs\\\\arcgis-mcp-env\\\\python.exe\",\n        \"ARCGIS_MCP_ALLOWED_ROOTS\": \"C:\\\\GIS\\\\Data;C:\\\\Workspace\",\n        \"ARCGIS_MCP_SCRATCH_GDB\": \"C:\\\\GIS\\\\Data\\\\scratch.gdb\",\n        \"ARCGIS_MCP_MAX_WORKERS\": \"2\"\n      }\n    }\n  }\n}\n```\n\nThis split-environment configuration assumes that the worker was provisioned\nfrom the repository with one of the Path B setup commands above. `PYTHONPATH`\nis not required when `uv sync --locked` has installed the project into the\nrepository `.venv`.\n\nA globally resolved `arcgis-mcp-server` command can work, but it creates a\nsplit-environment deployment. It is not recommended for first-time Windows\nsetup unless the worker environment has been provisioned and verified\nseparately.\n\nAfter restarting the MCP host, call `health_check` first. It verifies the\nserver-to-worker IPC path and reports the selected worker interpreter without\nimporting ArcPy. Then run a read-only ArcGIS tool or the ArcPy preflight command\nabove to validate the licensed runtime.\n\n---\n\n## 06 — Troubleshooting\n\n### `Worker process exited with code 1`\n\nIf the MCP server starts but every ArcGIS tool fails, inspect the server log for\nthe worker traceback. Common environment-related causes include:\n\n```text\nModuleNotFoundError: No module named 'arcgis_mcp'\nModuleNotFoundError: No module named 'pydantic_core._pydantic_core'\n```\n\nThese errors usually mean that the worker is using a different Python\ninstallation, the bridge was not installed into the worker interpreter, or the\nworker contains an incomplete or incompatible Pydantic installation. The recommended fix is the unified-environment configuration documented above.\n\nFor the recommended unified configuration, confirm that both values are\nidentical:\n\n```json\n\"command\": \"C:\\\\...\\\\envs\\\\arcgis-mcp-env\\\\python.exe\"\n```\n\n```json\n\"ARCPY_PYTHON_PATH\": \"C:\\\\...\\\\envs\\\\arcgis-mcp-env\\\\python.exe\"\n```\n\nInstall or update the bridge inside that exact interpreter:\n\n```powershell\n$ArcGISMcpPython = \"C:\\...\\envs\\arcgis-mcp-env\\python.exe\"\n& $ArcGISMcpPython -m pip install --upgrade arcgis-mcp-bridge\n```\n\nVerify the selected executable and bridge dependencies:\n\n```powershell\n& $ArcGISMcpPython -c \"import sys, arcgis_mcp, pydantic, pydantic_core; print(sys.executable); print('Bridge runtime OK')\"\n```\n\nThen verify ArcPy separately:\n\n```powershell\n& $ArcGISMcpPython -c \"import arcpy; print('ArcPy', arcpy.GetInstallInfo().get('Version'))\"\n```\n\nIf the bridge verification still fails specifically inside `pydantic_core`,\nreinstall Pydantic in the same environment so pip restores the matching\ncompiled dependency:\n\n```powershell\n& $ArcGISMcpPython -m pip install --upgrade --force-reinstall --no-cache-dir \"pydantic>=2.5,<3\"\n```\n\nRestart the MCP host completely after changing its Python environment or\nconfiguration.\n\n### `ARCPY_PYTHON_PATH` points to the wrong executable\n\nA Windows conda environment normally places its interpreter at the environment\nroot:\n\n```text\nC:\\...\\envs\\arcgis-mcp-env\\python.exe\n```\n\nDo not use another global Python installation or a nonexistent\n`arcgis-mcp-env\\Scripts\\python.exe` path.\n\n### Scratch geodatabase startup error\n\nThe default scratch workspace is:\n\n```text\n<first allowed root>\\scratch.gdb\n```\n\nIt must already exist. Create it in ArcGIS Pro or set\n`ARCGIS_MCP_SCRATCH_GDB` to an existing file geodatabase before restarting the\nMCP host.\n\n### `health_check` succeeds but ArcGIS tools still fail\n\n`health_check` intentionally verifies the server-to-worker process boundary\nwithout importing ArcPy. A successful result confirms IPC and interpreter\nselection, but it does not prove that ArcPy or an optional Esri extension\nlicense can be loaded.\n\nRun the ArcPy preflight command above and inspect the structured worker error\nfor `license`, `geoprocessing`, or `internal` details.\n\n---\n\n## 07 — Compatibility\n\n| ArcGIS Pro | Bundled Python | Status |\n|---|---|---|\n| 3.3 | 3.11 | ✅ Reference platform |\n| 3.4 | 3.11 | ⚠ Community-reported; verify with the preflight checks |\n| 3.1–3.2 | 3.9 | ❌ Unsupported by the current `Python >=3.11` package requirement |\n\n**Windows only.** ArcPy requires a licensed ArcGIS Pro installation on Windows.\nLayer A can run on other platforms for development and mocked CI, but Layer B\nrequires ArcGIS Pro.\n\nThe bridge package itself requires Python 3.11 or newer. ArcGIS Pro releases\nwhose cloned Python environment is older than 3.11 cannot run the current\nworker package.\n\n---\n\n## 08 — License\n\nApache License 2.0. See [LICENSE](LICENSE).\n\n<!-- mcp-name: io.github.muend/arcgis-mcp-bridge -->\n",
  "bytes": 24846,
  "sha": "0f4d09632072e9c361b97e113911d893a6fd3d6aed37bc49e1f2e54a72f38923",
  "repo_slug": "muend/arcgis-mcp-bridge",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/mcp_io_github_muend_arcgis_mcp_bridge_e3e09598/readme"
}