Skip to main content
Glama

Token Saver MCP

by jerry426
GitHub
TypeScript
MIT License
2
OverviewInspectNewSchemaRelated ServersReviewsScore
Need Help?View Source CodeReport Issue

Token Saver MCP - AI as a Full-Stack Developer

Transform AI from a code suggester into a true full-stack developer with complete control over both backend code and frontend browsers.

🆕 What's New in v1.0

Chrome DevTools Protocol Integration - AI can now:

  • 🌐 Control real browsers programmatically
  • 🧪 Test React components automatically
  • 🐛 Debug JavaScript errors in real-time
  • 📊 Analyze performance metrics
  • ✅ Verify code changes instantly

This isn't just an incremental improvement - it's a paradigm shift in AI-assisted development.

The Hidden Cost You Don't See

Every time your AI assistant searches your code, it's:

  • 🔥 Burning thousands of tokens on grep/find commands
  • ⏰ Making you wait 10-30 seconds for simple answers
  • 💸 Costing you money in API fees
  • 🔄 Repeatedly searching the same code

Meanwhile, VSCode's Language Server subsystem already knows everything about your code.

Token Saver MCP gives AI instant access to that knowledge. No searching. No waiting. No token waste.

What This Extension Does

Token Saver MCP provides TWO revolutionary toolsets that transform AI into a true full-stack developer:

🔍 Language Server Protocol (LSP) Tools

Gives AI direct access to VSCode's already-indexed code intelligence - delivering answers in milliseconds instead of seconds with 90% fewer tokens.

🌐 Chrome DevTools Protocol (CDP) Tools

Gives AI complete control over Chrome browsers for testing, debugging, and automation - enabling instant verification of code changes in real browsers.

Proven Results

  • ⚡ 100-1000x faster than text-based searching
  • 🎯 30+ production-ready tools (17 LSP + 13+ CDP)
  • 🛡️ Intelligent buffer protection prevents token overflow
  • 🔄 Zero configuration for single projects
  • 🚀 Full-stack capabilities - backend code + frontend browser

Real-World Performance

OperationTraditional ApproachWith Token Saver MCPYour Time Saved
Find where a function is defined5-10 seconds10ms☕ Make coffee
Find all usages of a variable10-30 seconds50ms💭 Keep your flow
Get type informationManual lookup5ms⚡ Instant
Rename across entire projectSeveral minutes100ms🚀 Already done
Search for text patterns2-15 seconds30ms📈 100x faster

Token & Cost Savings Calculator

MetricWithout Token SaverWith Token SaverYour Savings
Tokens per search~5,000 tokens~50 tokens99% fewer
Cost per search (GPT-4)$0.15$0.0015$0.1485
Daily searches (avg)5050-
Daily cost$7.50$0.075$7.43 saved
Monthly cost$225$2.25$222.75 saved
Yearly cost$2,700$27$2,673 saved

*Based on typical development patterns and AI code assistant pricing. Your savings may vary.

🌐 Revolutionary Browser Control (NEW!)

Token Saver MCP now includes Chrome DevTools Protocol integration, giving AI complete browser control:

Browser Testing & Automation Tools

ToolWhat It DoesExample Use
test_react_componentTest React components instantlyVerify component renders with correct props
test_api_endpointTest APIs from browser contextCheck CORS, auth, response times
test_form_validationTest forms completelySubmit, validate, check errors
check_page_performanceAnalyze performance metricsGet load times, memory usage, recommendations
debug_javascript_errorDebug JS errors in real-timeCapture stack traces, identify issues

Full-Stack Development Workflow

1. AI writes backend API endpoint (LSP tools) 2. AI launches browser automatically (CDP tools) 3. AI tests the API from frontend (CDP tools) 4. AI sees any errors immediately (CDP tools) 5. AI fixes the code (LSP tools) 6. AI verifies the fix works (CDP tools)

No more: "Please test this manually and tell me if it works" Now: AI tests everything itself, instantly!

Before & After Comparison

❌ Before Token Saver MCP - What You're Doing Now

AI: "Let me search for that function definition..." > Running: grep -r "functionName" . --include="*.ts" > [10 seconds pass...] > Found 47 matches, let me search more specifically... > Running: find . -name "*.ts" -exec grep -l "function functionName" {} \; > [15 more seconds...] > Tokens used: 5,000+ > Time wasted: 25 seconds > Result: Maybe found it, maybe not

✅ After Token Saver MCP - Instant Intelligence

AI: "Getting definition..." > Using: get_definition at file.ts:42:15 > [10ms] > Tokens used: 50 > Time: Instant > Result: Exact location with type info

🔬 Don't Take Our Word For It - Verify It Yourself!

Think these performance claims are too good to be true?

Don't trust us - let your own AI prove it! Install Token Saver MCP in your VSCode, then challenge your AI assistant to compare approaches:

Ask your AI to:

  1. Find a function definition - First with grep, then with get_definition
  2. Find all usages of a variable - First with text search, then with get_references
  3. Get type information - First by reading files, then with get_hover
  4. Understand code structure - First by parsing manually, then with get_document_symbols
  5. Trace function calls - First with grep patterns, then with get_call_hierarchy

Your AI will report:

  • 🎯 Exact token counts for each approach
  • ⏱️ Response time differences
  • 📊 Accuracy improvements
  • 💰 Cost savings per operation

The results speak for themselves - typically 100-1000x faster with 90-99% fewer tokens. Your AI assistant will confirm these aren't marketing claims - they're measurable facts.

Features

This extension exposes VSCode's Language Server Protocol features through MCP, providing AI assistants with 17 powerful tools:

Core Navigation & Intelligence

  • Go to Definition (get_definition) - Navigate to symbol definitions
  • Go to Type Definition (get_type_definition) - Navigate to type definitions
  • Find References (get_references) - Locate all usages of a symbol
  • Find Implementations (find_implementations) - Find all implementations of an interface/class
  • Hover Information (get_hover) - Access documentation and type information

Code Analysis & Refactoring

  • Code Completions (get_completions) - Get intelligent code suggestions at any position
  • Code Actions (get_code_actions) - Get available quick fixes and refactorings
  • Symbol Rename (rename_symbol) - Refactor symbols across the workspace
  • Document Symbols (get_document_symbols) - Get file structure with all symbols hierarchically
  • Call Hierarchy (get_call_hierarchy) - Trace incoming/outgoing function calls
  • Diagnostics (get_diagnostics) - Get errors, warnings, and hints for files
  • Text Search (search_text) - Search for text patterns across all files
  • Semantic Tokens (get_semantic_tokens) - Get detailed syntax highlighting information

Browser Control & Testing (CDP Tools)

  • Test React Component (test_react_component) - Complete React component testing
  • Test API Endpoint (test_api_endpoint) - Test APIs with CORS checks from browser
  • Test Form Validation (test_form_validation) - Automated form testing
  • Check Page Performance (check_page_performance) - Performance metrics & analysis
  • Debug JavaScript Error (debug_javascript_error) - Capture and analyze JS errors
  • Execute in Browser (execute_in_browser) - Run any JavaScript in browser context
  • Navigate Browser (navigate_browser) - Control browser navigation
  • Click/Type/Screenshot - Full browser automation capabilities

System & Utilities

  • Buffer Retrieval (retrieve_buffer) - Retrieve full data from buffered responses
  • Buffer Statistics (get_buffer_stats) - Monitor buffer system usage
  • Get Instructions (get_instructions) - Self-documenting API returns complete usage guide
  • Supported Languages (get_supported_languages) - Get all registered languages and their status

Intelligent Buffer System

The extension includes a sophisticated buffer system to prevent token overflow:

  • Automatic buffering for responses over 2,500 tokens (~10KB)
  • Smart previews with tool-specific intelligent summaries
  • Depth truncation to limit deeply nested data structures
  • Session-based isolation for multiple concurrent users
  • 60-second TTL with automatic cleanup

Performance Benefits

  • Instant results: <100ms response time vs 10-30 seconds with grep
  • Semantic accuracy: Real code intelligence, not text pattern matching
  • Zero additional cost: Leverages existing VSCode computation
  • Works across your workspace: Automatic Language Server activation for any file type - no manual setup needed

🚀 Get Started in 30 Seconds

# One command does everything: ./mcp setup /path/to/your/project

That's it! The extension automatically:

  • ✅ Finds an available port
  • ✅ Creates configuration files
  • ✅ Tests the connection
  • ✅ Provides the Claude command

Installation

VSCode Marketplace publishing is coming soon! Until then, here are your installation options:

  1. Download the latest .vsix file from the Releases page
  2. Install in VSCode using one of these methods:
    • Command Palette: Extensions: Install from VSIX... then select the file
    • Terminal: code --install-extension token-saver-mcp-*.vsix
    • Or drag and drop the .vsix file onto the VSCode Extensions view
  3. Reload VSCode to activate the extension

Option 2: Build from Source

  1. Clone and build the extension:
    git clone https://github.com/jerry426/token-saver-mcp.git cd token-saver-mcp pnpm install pnpm run build
  2. Create the extension package:
    pnpm exec vsce package --no-dependencies # This creates token-saver-mcp-*.vsix in the current directory # Note: --no-dependencies flag is required due to bundled dependencies
  3. Install the .vsix file in VSCode:
    code --install-extension token-saver-mcp-*.vsix
  4. Reload VSCode to activate the extension:
    • Command Palette: Developer: Reload Window
    • Or restart VSCode

Usage

🎯 IMPORTANT: For Best Results with AI Assistants

Tell your AI assistant to read the CLAUDE-MCP-USER.md file first!

"Please read the CLAUDE-MCP-USER.md file to understand how to use Token Saver MCP tools effectively"

This ensures your AI assistant:

  • Uses Token Saver MCP tools instead of slow text searches
  • Understands all 17 available tools and their parameters
  • Follows best practices for maximum performance
  • Saves 90-99% of tokens on code navigation tasks

The CLAUDE-MCP-USER.md file is the complete guide that teaches AI assistants how to leverage VSCode's Language Server intelligence instead of wasting tokens on grep/find operations.

Status Dashboard

See all your MCP-enabled projects:

./mcp status

Shows which projects are configured, running, and connected to Claude.

Multiple Projects Setup

Each project needs its own unique port number:

  1. Assign unique ports to each project:
    # Project A echo "9527" > /path/to/project-a/.lsp_mcp_port # Project B echo "9528" > /path/to/project-b/.lsp_mcp_port # Project C echo "9529" > /path/to/project-c/.lsp_mcp_port
  2. Set up Claude for each project:
    # Get the Claude command for each project ./mcp claude /path/to/project-a ./mcp claude /path/to/project-b
  3. The extension will automatically use the port from .lsp_mcp_port when you open each project

Port Discovery

To see all running MCP servers:

python3 test/find_mcp_servers.py

This shows which ports are actually in use and their workspace paths.

Available MCP Tools

Automatic Language Server Activation: The extension automatically activates the appropriate Language Server when you use any MCP tool. No manual file opening required - just call the tool and it works instantly! If a Language Server isn't already active, the extension will very briefly open and close a file of that type in the background to trigger activation. This happens automatically and takes just milliseconds.

Position-Based Tools

These tools require a file URI and position (line/character):

get_hover
{ "uri": "file:///path/to/file.ts", "line": 10, "character": 15 }

Returns: Type information, documentation, and signatures

get_completions
{ "uri": "file:///path/to/file.ts", "line": 10, "character": 15 }

Returns: Context-aware code completion suggestions

get_definition
{ "uri": "file:///path/to/file.ts", "line": 10, "character": 15 }

Returns: Location(s) where the symbol is defined

get_type_definition
{ "uri": "file:///path/to/file.ts", "line": 10, "character": 15 }

Returns: Location(s) where the type is defined

get_code_actions
{ "uri": "file:///path/to/file.ts", "line": 10, "character": 15 }

Returns: Available quick fixes and refactorings at the position

get_references
{ "uri": "file:///path/to/file.ts", "line": 10, "character": 15 }

Returns: All locations where the symbol is used

find_implementations
{ "uri": "file:///path/to/file.ts", "line": 10, "character": 15 }

Returns: All locations where the interface/class is implemented

get_call_hierarchy
{ "uri": "file:///path/to/file.ts", "line": 10, "character": 15, "direction": "incoming" // or "outgoing" }

Returns: Function call relationships - incoming shows callers, outgoing shows callees

rename_symbol
{ "uri": "file:///path/to/file.ts", "line": 10, "character": 15, "newName": "newSymbolName" }

Returns: Edit operations to rename across all files

Document-Level Tools

get_document_symbols
{ "uri": "file:///path/to/file.ts" }

Returns: Hierarchical structure of all symbols in the file (classes, methods, properties, etc.)

Search Tools

search_text
{ "query": "searchTerm", "useRegExp": false, "isCaseSensitive": false, "matchWholeWord": false, "maxResults": 100, "includes": ["**/*.ts"], "excludes": ["node_modules/**"] }

Returns: File locations and positions matching the search

Diagnostics Tools

get_diagnostics
{ "uri": "file:///path/to/file.ts" // Optional - if not provided, gets all diagnostics }

Returns: Errors, warnings, and hints for the file(s)

get_semantic_tokens
{ "uri": "file:///path/to/file.ts" }

Returns: Detailed syntax highlighting information with token types and modifiers

Buffer Management Tools

retrieve_buffer
{ "bufferId": "get_document_symbols_1754955026362_z2ksv6t8z" }

Returns: Complete original data from a buffered response

get_buffer_stats
{} // No parameters required

Returns: Active buffer count, total size, oldest buffer age

Self-Documentation Tool

get_instructions
{} // No parameters required

Returns: Complete CLAUDE-MCP-USER.md guide with all usage instructions

get_supported_languages
{} // No parameters required

Returns: All languages registered in VSCode organized by category, active languages in workspace, and total count

Configuration

VSCode settings:

  • lsp-mcp.enabled - Enable/disable the MCP server (default: true)
  • lsp-mcp.port - Server port (default: 9527)
  • lsp-mcp.maxRetries - Port retry attempts if occupied (default: 10)

Testing

Test all MCP tools:

python3 test/test_mcp_tools.py

Expected output:

✓ Hover works - returned documentation ✓ Completions works - returned 193 items ✓ Definition works - found 1 location(s) ✓ References works - found 5 reference(s) ✓ Implementations works - interface/class implementations ✓ Document symbols works - found 10 top-level symbol(s) ✓ Text search works - found 6 match(es) ✓ Call hierarchy works - target: startMcp, 2 call(s) ✓ Rename works - would affect 2 file(s) ✓ Type definition works - found 1 location(s) ✓ Code actions works - found 3 quick fix(es) ✓ Diagnostics works - found issues in 2 file(s) ✓ Semantic tokens works - 500+ tokens decoded ✓ Buffer system works - response buffered (8744 tokens) ✓ retrieve_buffer works - retrieved 21 items ✓ Buffer stats works - 3 active buffer(s) 🎉 SUCCESS! All 17 tools are working!

Architecture

┌─────────────┐ MCP/HTTP ┌──────────────┐ │ AI Assistant│ ◄──────────────► │ MCP Server │ │ (e.g. Claude)│ │ (Port 9527) │ └─────────────┘ └──────────────┘ │ ▼ ┌──────────────┐ │ LSP Bridge │ │ (TypeScript)│ └──────────────┘ │ ▼ ┌──────────────┐ │ VSCode LSP │ │ APIs │ └──────────────┘

Key design principles:

  • Automatic Language Server activation: The extension automatically activates Language Servers as needed
  • Workspace-wide access: All files are immediately accessible without manual activation
  • Session management: Each MCP client gets an isolated session
  • Consistent formatting: All responses return JSON-serialized data
  • Error resilience: Graceful handling of missing files or invalid positions

Development

# Install dependencies pnpm install # Development with watch mode pnpm run dev # Build for production pnpm run build # Run tests pnpm run test # Lint code pnpm run lint # Type checking pnpm run typecheck

Project Structure

src/ ├── index.ts # Extension entry point ├── mcp/ │ ├── index.ts # MCP server implementation │ ├── tools.ts # Tool registrations │ └── buffer-manager.ts # Intelligent buffer system ├── lsp/ │ ├── hover.ts # Hover information │ ├── completion.ts # Code completions │ ├── definition.ts # Go to definition │ ├── references.ts # Find references │ ├── implementations.ts # Find implementations │ ├── document-symbols.ts # File structure │ ├── call-hierarchy.ts # Call tracing │ ├── rename.ts # Symbol rename │ └── text-search.ts # Text search └── utils/ └── index.ts # Logging utilities

Troubleshooting

Extension not responding?

# Check if MCP server is running python3 test/find_mcp_servers.py

Need a specific port?

echo "9527" > .lsp_mcp_port

Port already in use?

  • Check .lsp_mcp_port file or let extension auto-increment from default port

MCP tools not responding?

  • Ensure VSCode has the workspace open
  • Language Servers are automatically activated when you use any MCP tool
  • Check extension is enabled in settings (lsp-mcp.enabled)
  • Verify the MCP server is running on the expected port

Claude Code stuck on "connecting..." or showing "failed"?

  • Reload the VSCode window with Cmd+Shift+R (Mac) or Ctrl+Shift+R (Windows/Linux) to restart the extension and MCP server
  • This clears any stale connection states and forces a fresh initialization
  • After reloading, restart Claude Code in your project terminal

Large responses causing issues?

  • Responses over 2,500 tokens are automatically buffered
  • Use the returned bufferId with retrieve_buffer tool to get full data
  • Check buffer stats with get_buffer_stats tool

Testing the connection:

# Run the comprehensive test suite python3 test/test_mcp_tools.py

Contributing

Contributions welcome! Please check the roadmap above for features to implement.

License

MIT

This server cannot be installed

-
security - not tested
A
license - permissive license
-
quality - not tested

How are these scores calculated?

local-only server

The server can only run on the client's local machine because it depends on local resources.

Bridges VSCode's Language Server Protocol with MCP to give AI assistants instant access to code intelligence, delivering 100-1000x faster responses with 90% fewer tokens than traditional text-based searching. Provides 17 production-ready tools for navigation, refactoring, diagnostics, and code analysis.

  1. 🆕 What's New in v1.0
    1. The Hidden Cost You Don't See
      1. What This Extension Does
        1. 🔍 Language Server Protocol (LSP) Tools
        2. 🌐 Chrome DevTools Protocol (CDP) Tools
      2. Proven Results
        1. Real-World Performance
          1. Token & Cost Savings Calculator
            1. 🌐 Revolutionary Browser Control (NEW!)
              1. Browser Testing & Automation Tools
              2. Full-Stack Development Workflow
            2. Before & After Comparison
              1. ❌ Before Token Saver MCP - What You're Doing Now
              2. ✅ After Token Saver MCP - Instant Intelligence
            3. 🔬 Don't Take Our Word For It - Verify It Yourself!
              1. Features
                1. Core Navigation & Intelligence
                2. Code Analysis & Refactoring
                3. Diagnostics & Search
                4. Browser Control & Testing (CDP Tools)
                5. System & Utilities
                6. Intelligent Buffer System
                7. Performance Benefits
              2. 🚀 Get Started in 30 Seconds
                1. Installation
                  1. Option 1: Install from Pre-built Release (Recommended)
                  2. Option 2: Build from Source
                2. Usage
                  1. 🎯 IMPORTANT: For Best Results with AI Assistants
                  2. Status Dashboard
                  3. Multiple Projects Setup
                  4. Port Discovery
                3. Available MCP Tools
                  1. Position-Based Tools
                  2. Document-Level Tools
                  3. Search Tools
                  4. Diagnostics Tools
                  5. Buffer Management Tools
                  6. Self-Documentation Tool
                4. Configuration
                  1. Testing
                    1. Architecture
                      1. Development
                        1. Project Structure
                      2. Troubleshooting
                        1. Contributing
                          1. License

                            Related MCP Servers

                            • code2prompt-mcp

                              ODAncona
                              -
                              security
                              A
                              license
                              -
                              quality
                              An MCP server that analyzes codebases and generates contextual prompts, making it easier for AI assistants to understand and work with code repositories.
                              Last updated -
                              10
                              Python
                              MIT License

                            • Vibe Coder MCP

                              freshtechbro
                              A
                              security
                              F
                              license
                              A
                              quality
                              An MCP server that supercharges AI assistants with powerful tools for software development, enabling research, planning, code generation, and project scaffolding through natural language interaction.
                              Last updated -
                              11
                              668
                              69
                              TypeScript
                              • Linux
                              • Apple

                            • Kodit

                              helixml
                              -
                              security
                              A
                              license
                              -
                              quality
                              A Code Indexing MCP Server that connects AI coding assistants to external codebases, providing accurate and up-to-date code snippets to reduce mistakes and hallucinations.
                              Last updated -
                              75
                              Python
                              Apache 2.0

                            • Moatless MCP Server

                              TwT23333
                              A
                              security
                              F
                              license
                              A
                              quality
                              A Model Context Protocol (MCP) server for advanced code analysis and editing with semantic search capabilities, enabling AI assistants to perform complex code operations through a standardized interface.
                              Last updated -
                              15
                              1
                              Python

                            View all related MCP servers

                            New MCP Servers

                            MCP directory API

                            We provide all the information about MCP servers via our MCP API.

                            curl -X GET 'https://glama.ai/api/mcp/v1/servers/jerry426/token-saver-mcp'

                            If you have feedback or need assistance with the MCP directory API, please join our Discord server