Standalone inspector for MCP servers with UI preview and LLM-assisted tool orchestration.

Version: 1.3.6

Features

• multi-server MCP support (HTTP and stdio)
• tools, resources, and prompts explorer
• MCP Apps UI preview with tool-call bridge
• chat playground with independent multi-tab sessions
• LLM orchestration: Ollama, OpenAI, Anthropic
• settings panel for Inspector/OpenAI/Claude API keys
• auto-tools routing (LLM decides if tools are needed)
• test suites, metrics, logs, export
• secure mode: API key, CORS allowlist, rate limiting

Desktop Install (Recommended)

For normal users, install from GitHub Releases:

• Windows: PolyMCP Inspector_*_x64-setup.exe (or .msi)
• Ubuntu: .deb package (or AppImage if published)

No Python, Node, or manual build steps are required for end users.

Python Install (Source/Dev)

pip install -e .

Run (Source/Dev)

polymcp-inspector --host 127.0.0.1 --port 6274 --no-browser

Open:

http://127.0.0.1:6274/

Secure Mode

polymcp-inspector --secure
# or
polymcp-inspector --secure --api-key "your-strong-key"

LLM Providers

Set credentials using either method:

1. Settings tab in the Inspector UI (recommended):
   • Inspector API key (for secure mode)
   • OpenAI API key
   • Claude (Anthropic) API key
   • Keys are stored in browser localStorage for that Inspector UI
2. Environment variables:

export OLLAMA_BASE_URL=http://127.0.0.1:11434
export OPENAI_API_KEY=...
export ANTHROPIC_API_KEY=...

The chat playground lets you pick provider and model per tab. Each tab keeps its own provider/model/server selection and conversation history.

When keys are configured in Settings, requests include X-OpenAI-API-Key and X-Anthropic-API-Key headers. Backend env vars are used as fallback.

URL query support:

• ?api_key=...
• ?openai_api_key=...
• ?anthropic_api_key=...

MCP Apps UI

If your server exposes HTML resources (app://...), the Inspector renders them inside chat as part of the assistant response. The latest UI is live and previous responses are snapshotted.

Desktop App (Tauri)

This repo includes a desktop wrapper in desktop-tauri/.

Build steps (for maintainers):

1. Build backend sidecar exe (Windows):
   • powershell -ExecutionPolicy Bypass -File .\scripts\build_backend_exe.ps1
   • (Optional) choose Python explicitly: powershell -ExecutionPolicy Bypass -File .\scripts\build_backend_exe.ps1 -PythonExe C:\Python312\python.exe
   • TS backend sidecar alternative: powershell -ExecutionPolicy Bypass -File .\scripts\build_backend_ts_exe.ps1
   • (Optional) custom pkg targets: powershell -ExecutionPolicy Bypass -File .\scripts\build_backend_ts_exe.ps1 -PkgTargets "node18-win-x64,node16-win-x64"
2. Build installer:
   • cd desktop-tauri
   • npm install
   • npm run tauri:build
3. Publish installer artifacts:
   • desktop-tauri/src-tauri/target/release/bundle/nsis/*.exe
   • desktop-tauri/src-tauri/target/release/bundle/msi/*.msi

TS Backend (Parity Migration)

The repo now includes an Inspector backend in TypeScript under ts-backend/ with API-compatible routes used by the current UI.

Build and run backend:

cd ts-backend
npm install
npm run build
node dist/server.js --host 127.0.0.1 --port 6274 --secure --api-key test --verbose

To make desktop Tauri prefer TS backend in local/dev runs:

set POLYMCP_INSPECTOR_BACKEND_MODE=ts