Test, debug, and develop Model Context Protocol servers with a beautiful glassmorphism UI and a powerful Workspace mode. Record scenarios • Collections + run reports • Workspace canvas • Workflow export (Python + Node) • • •
A powerful MCP testing tool with multi-provider LLM support (Ollama, OpenAI, Claude, Gemini). Test, debug, and develop MCP servers with a modern UI.
ดาว
18
เครื่องมือ
—
อัปไทม์
—
Mock servers
Docs generator
Zero config with Ollama
Need help? Use the Studio Assistant (bottom‑right compass) to learn features, generate workflows, import OpenAPI, and navigate the UI — just ask it.
🆕 Major Update (January 2025): This version includes significant new features like OAuth config UI, LLM settings persistence, E2E tests, session management, security enhancements, and much more. The previous stable version is available in the v2 branch as a backup.
🎬 Showcase
Workspace Canvas
Studio Assistant
AI Workflow Builder
Documentation Generator
📌 Local MCP Test Bench - Designed for development. Not intended for internet exposure.
⚡ Get Started in 60 Seconds
# 1. Clone and install
git clone https://github.com/JoeCastrom/mcp-chat-studio.git && cd mcp-chat-studio && npm install
# 2. Start (works with Ollama out of the box - no API keys needed!)
npm run dev
# 3. Open http://localhost:3082 and start testing! 🎉
That's it! You now have a complete MCP testing platform running locally.
Toggle Classic / Workspace in the header to pick your workflow.
Use the model badge in the header to switch LLM providers; the ⚙️ Settings button handles model/auth tuning.
🆚 Why Not Just Use Claude Desktop or ChatGPT?
Feature
MCP Chat Studio
Claude Desktop
ChatGPT
Test ALL tools automatically
✅
❌
❌
Record & Replay scenarios
✅
❌
❌
Workflow debugger (breakpoints)
✅
❌
❌
Bulk testing (multiple inputs)
✅
❌
❌
Response diffing
✅
❌
❌
Contract testing
✅
❌
❌
Tool usage analytics
✅
❌
❌
Custom assertions (14 operators)
✅
❌
❌
Low-level MCP debugging
✅
❌
❌
Mock server generator
✅
❌
❌
9 LLM providers + custom
✅
❌ (Claude only)
❌ (OpenAI only)
Local/no API keys needed
✅ (Ollama)
❌
❌
Multi-environment profiles
✅
❌
❌
Session branching
✅
❌
❌
🧭 Two Layouts: Classic + Workspace
Classic Mode keeps the familiar sidebar + tabs for focused work. Workspace Mode turns every tool into a draggable panel on an infinite canvas:
Radial menu to add panels (right-click)
Quick access bar for fast panel focus
Zoom, pan, mini-map, fit-all
Command palette (Ctrl+K / Ctrl+Shift+P)
Workspace sessions + export/import bundles
🎨 Highlights: Advanced Testing & Debugging Suite
🐛 Workflow Debugger
Debug complex workflows with professional debugging tools:
Breakpoint debugging - Pause execution at any node
Step-through mode - Execute one node at a time
Variable inspection - View inputs, outputs, and context
Session management - Pause, resume, step, or abort
✨ AI Workflow Builder
Generate workflows from a natural-language goal:
Tool-aware prompts - Uses connected MCP tools as context
One-click generate - Creates a workflow you can edit and run
Works with Workspace and Classic modes
Preflight validation - Catch missing tool args and invalid JSON before execution
📦 Workflow Export
Turn visual flows into runnable scripts:
Python (mcp SDK) and Node.js (mcp SDK)
Tool calls + variable substitution baked in
Shareable automation for CI or docs
🔍 Advanced Inspector
Three powerful new testing tools:
📊 Timeline - Chronological log of all JSON-RPC messages
🧪 Bulk Test - Execute tools with multiple inputs (parallel/sequential)
🔥 Bulk Test Heatmap - Visualize latency spread across inputs + export JSON with heatmap data
🎬 Fuzz Fail → Scenario - Save failing cases as replayable scenarios
📚 Failure Datasets - Store bulk-test failures as dataset rows for data runs
🔀 Diff - Side-by-side comparison with similarity scoring
🌐 Cross-Server Snapshot - Run one tool across servers and diff output
🧠 Ollama model picker - UI dropdown of locally installed models (auto-detected)
🔄 Provider Switcher - Swap LLMs from the model badge and manage visible providers in one place
🧭 Studio Assistant - Floating help chat aware of your current panel/layout, with dock + pop‑out modes, quick actions, FAQ fallback, and drag‑and‑drop OpenAPI import
Run collections from the command line for CI/CD integration:
# Run a collection
mcp-test run ./collections/my-collection.json --server http://localhost:3082
# Run with iteration data
mcp-test run ./collections/my-collection.json --data ./datasets/users.json
# JUnit output for CI
mcp-test run ./collections/my-collection.json --reporters junit --export ./reports/junit.xml
# Authenticated run (OAuth/session-based)
mcp-test run ./collections/my-collection.json --session <session-id>
🧬 Schema Regression CI
Capture tool schemas and gate changes in CI:
# Snapshot current tool schemas
mcp-test schema snapshot --out ./schema-baseline.json
# Compare snapshots and fail on change
mcp-test schema diff ./schema-baseline.json ./schema-current.json --format junit --out ./schema-diff.xml --gate
In the UI, use Schema Watch under Contracts to auto-detect drift during live testing.
🚦 Collection Run Gates (CI)
Export a Gate file from any Collection Run Report and fail CI on regressions:
# From the UI: click 🚦 Export Gate to download collection-run-gate.json
node scripts/collection-gate.js ./collection-run-gate.json
🎙️ Mock Recorder (Live Capture)
Record live tool calls and instantly spin them into a mock server:
Capture tool calls from the Inspector
Create mock with one click
Keep context with real tool schemas and outputs
🌿 Session Branching
Fork conversations - Branch at any message
Save snapshots - Preserve conversation states
Load branches - Continue from any saved point
Branch manager - View, load, delete branches
🌍 Multi-Environment Profiles
Dev/Staging/Prod - Switch between environments
Auto-save config - Per-environment settings
Global + env variables - Shared and environment-scoped variables
Quick switching - Sidebar dropdown
📦 Installation
Prerequisites
Node.js 18+
npm or yarn
Quick Start
# Clone the repository
git clone https://github.com/JoeCastrom/mcp-chat-studio.git
cd mcp-chat-studio
# Install dependencies
npm install
# Copy environment template
cp .env.example .env
# Start the server
npm run dev
In-App Quick Start
Use the 🚀 Start button in the header to open a guided checklist that jumps you into Inspector, Scenarios, Collections, Contracts, and the Generator.
If you already have Ollama installed on your machine:
# Start MCP Chat Studio only (uses your local Ollama)
docker-compose up
# Run in background
docker-compose up -d
The app automatically connects to your local Ollama at http://localhost:11434.
Option 1b: Lite Profile (No Python/MCP Auto-Run)
If you want a smaller container without Generator auto‑run:
docker-compose --profile lite up mcp-chat-studio-lite
Option 2: Include Ollama in Docker
If you don't have Ollama installed locally:
# Start both MCP Chat Studio AND Ollama in Docker
docker-compose --profile with-ollama up
# Run in background
docker-compose --profile with-ollama up -d
# Create .env file or set environment variables
OPENAI_API_KEY=sk-your-key
ANTHROPIC_API_KEY=sk-ant-your-key
GOOGLE_API_KEY=your-google-key
Using Docker Only
# Build image
docker build -t mcp-chat-studio .
# Run with Ollama (default)
docker run -p 3082:3082 mcp-chat-studio
# Run with OpenAI
docker run -p 3082:3082 \
-e OPENAI_API_KEY=sk-your-key \
mcp-chat-studio
# Run with custom config
docker run -p 3082:3082 \
-v $(pwd)/config.yaml:/app/config.yaml:ro \
mcp-chat-studio
Lite Docker Image (Smaller)
If you don’t need Generator auto‑run, use the lite image:
# Build lite image
docker build -f Dockerfile.lite -t mcp-chat-studio:lite .
# Run lite image
docker run -p 3082:3082 mcp-chat-studio:lite
Note: Lite image omits Python/MCP, so “Run & Connect (Auto)” is disabled.
Persisting Studio Data (Recommended)
Persist sessions, mocks, monitors, OAuth tokens, and logs by mounting data/:
docker run -p 3082:3082 \
-v $(pwd)/data:/app/data \
mcp-chat-studio
If you use Generator → Run & Connect (Auto) in Docker, also mount the temp project folder:
MCP Chat Studio supports 9 LLM providers + Custom. Configure in config.yaml or from the UI (⚙️):
UI changes are saved locally in data/llm-config.json (including optional auth settings).
API keys can be entered directly in the UI (they are persisted in the same file). .env is still supported for headless/CI setups.
Provider Visibility (optional)
Hide providers per-user in the UI, or enforce a server allow-list:
# Only show/allow these providers in the UI
LLM_ALLOWED_PROVIDERS=ollama,openai,custom
Custom endpoints also work with Bearer tokens (set auth.type: bearer and provide api_key).
In the UI, we’ll auto‑prefix Bearer if you paste a raw token.
If your gateway needs an extra header, set it in ⚙️ LLM Settings → Extra Auth Header.
OpenRouter (100+ Models)
Access Claude, GPT-4, Gemini, Llama, and 100+ more models through a single API:
llm:
provider: openrouter
model: anthropic/claude-3.5-sonnet # or openai/gpt-4o, google/gemini-pro-1.5
# .env
OPENROUTER_API_KEY=your-key
# Get your key at https://openrouter.ai/keys
OAUTH_CLIENT_ID=your-client-id
OAUTH_CLIENT_SECRET=your-secret
OAUTH_AUTHORIZE_URL=https://... # For custom providers
OAUTH_TOKEN_URL=https://...
OAUTH_DISABLE_SSL_VERIFY=true # Dev only (self-signed/internal PKI)
MCP_SANDBOX_ENGINE=vm2 # Optional: set to isolated-vm if installed
Click Login to authenticate.
🔐 Security Best Practices
Security Environment Variables
Variable
Default
Description
ENABLE_SCRIPTS
false
Enable script execution (ScriptRunner, JS workflow nodes). Disabled by default for security.
HOST
127.0.0.1
Server bind address. Defaults to localhost only.
ALLOW_REMOTE
false
Required if HOST=0.0.0.0. Prevents accidental network exposure.
CORS_ORIGINS
localhost
Comma-separated list of allowed origins for CORS.
Script Execution
Script execution (pre/post-request scripts, JavaScript workflow nodes) is disabled by default for security. To enable:
ENABLE_SCRIPTS=true npm run dev
Network Binding
By default, the server binds to 127.0.0.1 (localhost only). To expose to all network interfaces:
HOST=0.0.0.0 ALLOW_REMOTE=true npm run dev
The server will refuse to start if HOST=0.0.0.0 without ALLOW_REMOTE=true to prevent accidental exposure.
Important Security Notes
⚠️ Never commit .env file - Contains API keys and secrets
⚠️ Use environment variables - All secrets should be in .env or environment
⚠️ SSL verification enabled - Disable only for dev/self-signed certs (UI toggle or OAUTH_DISABLE_SSL_VERIFY=true)
🔐 OAuth token storage - Set OAUTH_TOKEN_KEY to persist tokens encrypted in data/oauth-tokens.json (memory-only if unset). Use Redis/DB for multi-user production.
⚠️ Sandbox engine - Defaults to isolated-vm when installed (falls back to vm2). Set MCP_SANDBOX_ENGINE to override.
🔑 LLM API keys (UI) - Saved server-side in data/llm-config.json when entered in ⚙️ LLM Settings (not localStorage). For custom Bearer auth, the UI auto‑prefixes Bearer if you paste a raw token. Use .env for headless/CI.
🔒 CSRF protection - Browser requests require X-CSRF-Token (UI adds it automatically). CLI requests without an Origin header are allowed.
📝 Audit logging - Security-relevant events are written to data/audit.log.
💾 Server-side sessions - Chat sessions/tool history sync to data/sessions.json (tied to sessionId cookie).
🧼 HTML sanitization - DOMPurify is used when installed (fallback allowlist sanitizer).
🔗 Session share links - Share the current session via Settings → "Share" (one-time link to import a session snapshot).
For Development
# .env (local development)
OPENAI_API_KEY=sk-your-key
ANTHROPIC_API_KEY=sk-ant-your-key
# Disable SSL verify only for internal/self-signed certs
OAUTH_DISABLE_SSL_VERIFY=true
For Production
# Use secure environment variables
export OPENAI_API_KEY="sk-your-key"
export ANTHROPIC_API_KEY="sk-ant-your-key"
# Never disable SSL verification in production
# OAUTH_DISABLE_SSL_VERIFY should NOT be set
# Use Redis for OAuth tokens (multi-instance deployments)
# Modify server/services/OAuthManager.js to use Redis instead of Map
MCP Server Security
Filesystem Server: Only grant access to specific directories
Database Servers: Use read-only credentials when possible
API Keys: Store in .env, never in config.yaml
Network Servers (SSE): Use HTTPS and authentication
What's Safe to Commit
✅ Safe:
config.yaml.example
.env.example
Source code
Documentation
❌ NEVER Commit:
.env (actual credentials)
config.yaml (may contain secrets)
API keys or tokens
Certificates or private keys
❓ Frequently Asked Questions
General
Q: Can I use multiple MCP servers at once?
A: Yes! Add as many as you want. Tools are automatically namespaced (e.g., github__get_issue, filesystem__read_file) to prevent conflicts.
Q: Do I need API keys to start?
A: No! Use Ollama locally (free, no keys required). API keys are only needed for cloud LLM providers like OpenAI, Anthropic, etc.
Q: Can I use my own MCP server?
A: Absolutely! Add it via the UI or edit config.yaml. Supports both STDIO and SSE transports.
Q: Does this work on Windows?
A: Yes! Works on Windows, macOS, and Linux. For Windows, use PowerShell or Git Bash for commands.
Usage
Q: How do I test tools without the chat?
A: Use the Inspector tab to call tools directly without involving the LLM. Perfect for debugging and testing.
Q: Why can't I use streaming with tools?
A: LLMs need the complete response to decide which tools to call. Streaming is auto-disabled when tools are enabled.
Q: What are "risky tools"?
A: Tools that have side effects (clicking, typing, launching apps). They're skipped by default in "Test All Tools" to prevent unwanted actions.
Q: How do I switch LLM providers?
A: Click ⚙️ Settings → Change provider and model → Save. Changes take effect immediately.
MCP Servers
Q: What's the difference between STDIO and SSE?
A: STDIO runs locally as a subprocess. SSE connects to remote servers over HTTP. Use STDIO for local tools, SSE for network services.
Q: Can MCP servers see my API keys?
A: Only if you explicitly pass them via environment variables. Each server only sees the env vars you configure for it.
Q: My MCP server won't connect. What's wrong?
A: Check:
Q: Can I use the same MCP server with different configurations?
A: Yes! Add it multiple times with different names and configs. For example, github-personal and github-work with different tokens.
Development
Q: How do I add a new LLM provider?
A: Edit server/services/LLMClient.js. See CONTRIBUTING.md for details.
Q: Can I contribute new features?
A: Yes! We welcome contributions. See CONTRIBUTING.md for guidelines.
Q: Is there an API for automation?
A: Yes! See the API Endpoints section. All features are accessible via REST API.
🐛 Troubleshooting
MCP Server Won't Connect
Check the command/path is correct
Verify environment variables
Check server logs in console
Try running the command manually
LLM Not Responding
Check provider settings in ⚙️ Settings
Verify Ollama is running (ollama serve)
Check API key for OpenAI
Tools Show ⚠️ Warning
This means the tool responded but our dummy test input was invalid. The tool is working - it just needs proper arguments.
📜 License
MIT License - feel free to use and modify.
🤝 Contributing
We welcome contributions from the community! Whether it's bug fixes, new features, documentation improvements, or examples - all contributions are appreciated.
How to Contribute
Fork the repository
Clone your fork: git clone https://github.com/YOUR_USERNAME/mcp-chat-studio.git
Create a branch: git checkout -b feature/amazing-feature
คัดลอกคอนฟิก stdio สำหรับ Cursor, Claude Desktop, Claude Code หรือ VS Code ได้ในหน้านี้ คอนฟิกนี้ใช้การติดตั้งของผู้เผยแพร่ ไม่ใช่พร็อกซีของ PlugMCP
MCP Chat Studio ต้องใช้ API key ไหม?
ระเบียนนี้ไม่ได้ระบุตัวแปรสภาพแวดล้อมที่จำเป็น
MCP Chat Studio ใช้ทรานสปอร์ตแบบใด?
MCP Chat Studio ทำงานแบบโลคัลผ่าน stdio
MCP Chat Studio ถูกตรวจสอบล่าสุดเมื่อใด?
ตรวจสอบล่าสุดเมื่อ 2026-10-28 A powerful MCP testing tool with multi-provider LLM support (Ollama, OpenAI, Claude, Gemini). Test, debug, and develop MCP servers with a modern UI.