extra_skills_mcp_tools
Enables generation of chart specifications compatible with Ant Design's chart library, supporting line, bar, pie, scatter, and histogram chart types.
Provides tools for repository information, README content, issue tracking, commit history, repository search, and file content retrieval via the GitHub API.
Supports data manipulation with pandas, including creating DataFrames, filtering, aggregation, correlation analysis, and outlier detection.
Allows web, image, and news searches through a SearXNG instance, with status checking capability.
Offers external storage and retrieval of tool outputs using SQLite, enabling token reduction by storing large results outside the context.
Provides tools for fetching video transcripts (with timestamp options), search, video metadata, batch transcription, and summarization.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@extra_skills_mcp_toolssearch the web for latest AI research papers"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
MCP Server Suite — Token-Optimized AI Development Platform
Self-hosted MCP server with 83+ tools for AI workflows — featuring 80%+ token reduction through 5 optimization layers.
Table of Contents
Related MCP server: token-pilot
Overview
This MCP server provides 83+ tools for AI-powered development workflows, featuring:
🌐 Web Search & Scraping — SearXNG, Firecrawl, structured extraction
🐙 GitHub Integration — Repos, issues, commits, search
📺 YouTube — Transcripts, search, summarization
💻 Code Execution — Python sandbox, testing
📊 Data & Charts — Pandas, matplotlib visualization
🧠 Engineering Intelligence — Task classification, bug tracing, memory
⚡ Token Optimization — Built-in 80% token reduction
No API keys required for core functionality.
Quick Start
# 1. Clone repository
git clone https://github.com/simpletoolsindia/extra_skills_mcp_tools.git
cd extra_skills_mcp_tools
# 2. Run installation script
./install-claude-code.sh
# 3. Restart Claude Code
claude
# 4. Verify installation
docker compose -f docker-compose.local.yml psToken Optimization (80% Savings)
This server implements 5 layers of token optimization to maximize your context window:
Optimization Layers
Layer | Reduction | Description |
Tool Trimming | 80% | 90 → 64 tools with concise descriptions |
Web Content | 80-97% | Clean markdown, strip nav/ads/scripts |
Context Mode | 98% | External SQLite storage for outputs |
Lazy Loading | 91% | Load schemas on-demand |
Semantic Search | 91% | Natural language tool discovery |
Token Comparison
Metric | Before | After | Savings |
Tool List | ~13,500 tokens | ~2,700 tokens | 80% |
Web Fetch | ~8,000 tokens | ~2,000 tokens | 75% |
Tool Output | ~5,000 tokens | ~50 tokens | 98% |
Full Workflow | ~200,000 tokens | ~30,000 tokens | 85% |
Optimization Tools
Tool | Usage | Tokens |
| Ultra-fast title + summary | ~25-300 |
| Clean markdown with tracking | ~2,000 |
| Article/product/table | ~1,500 |
| Store output externally | ~50 |
| List without full schemas | ~2,000 |
| "search the web" → searxng_search | ~500 |
Usage Examples
# Before: Fetch raw HTML (~8000 tokens)
fetch("https://example.com") # Returns bloated HTML
# After: Optimized fetch (~300 tokens)
quick_fetch(url="https://example.com", max_tokens=1500)
# Returns: {title: "Example", summary: "...", tokens: 300}
# Store large outputs externally (98% reduction)
ctx_store_output(
tool_name="github_repo",
arguments={"owner": "anthropics"},
output={"repo": "claude-code", "stars": 15000}
)
# Returns: {"ref": "@ctx:default:abc123", "size_bytes": 54}
# Instead of storing 500+ tokens, just store the referenceClaude Code Optimization
Model Selection Strategy
Use the right model for the right task:
Model | Best For | Cost | When to Use |
Sonnet | Most coding | $3/1M tokens | Default choice, ~60% cheaper |
Haiku | Code review, docs | $0.25/1M tokens | Routine tasks, fixes |
Opus | Complex refactoring | $15/1M tokens | Architecture, deep debugging |
Recommended Settings
Add to ~/.zshrc or ~/.bashrc:
# Model Settings
claude config set --claude-code-subagent-model sonnet
export HAIKU_MODEL=haiku
# Thinking Token Limit (~70% savings)
# Default: 32,000 tokens
# Recommended: 10,000 tokens
export MAX_THINKING_TOKENS=10000
# Compaction Settings (better performance)
# Default: 95% context before compact
# Recommended: 50% for more working room
export CLAUDE_AUTOCOMPACT_PCT_OVERRIDE=50
# MCP Server Limit
# Keep under 10 MCP servers, 80 total tools
# More = reduced effective context windowQuick Commands
Command | Purpose |
| Monitor token usage and costs |
| Free context reset between tasks |
| Manual compaction at breakpoints |
| Check current context usage |
MCP Server Best Practices
Warning: Each MCP server adds tool definitions to context. Excessive servers reduce effective context from 200k to ~70k.
Recommended: Keep under 10 MCP servers, 80 total tools.
10 Strategies to Reduce MCP Token Bloat
Design tools with intent — Single purpose, clear inputs/outputs
Cache aggressively — Identical queries hit cache
Minimize server usage at runtime — Enable only when needed
Group tools by domain — Logical grouping reduces confusion
Deploy subagents — Route routine tasks to Haiku
Just-in-time context loading — Load schemas on-demand
Externalize computational results — Store large outputs
Apply advanced data filtering — Filter at extraction time
Externalize cross-cutting concerns — Centralize auth/errors
Keep tools lean — Runtime handles concerns centrally
Cost Comparison
Scenario | Before | After | Savings |
1 hour coding | $2.50 | $0.75 | 70% |
1 day research | $8.00 | $2.40 | 70% |
1 week project | $35.00 | $10.50 | 70% |
All Tools (100+)
🌐 Web Search & Scraping (10)
Tool | Description |
| Web search via SearXNG (Pi5) |
| Image search |
| News search |
| Check SearXNG status |
| Clean markdown extraction |
| Article/product/table extraction |
| Ultra-fast title + summary |
| JavaScript-heavy pages (Playwright) |
| Advanced scraping |
| Article extraction |
🐙 GitHub (6)
Tool | Description |
| Repository information |
| README content |
| List issues |
| List commits |
| Search repositories |
| Get file content |
📺 YouTube (6)
Tool | Description |
| Get transcript |
| Timestamped transcript |
| Search videos |
| Video metadata |
| Batch transcription |
| Summarize transcript |
📰 Hacker News (6)
Tool | Description |
| Top stories |
| Newest stories |
| Best stories |
| Ask HN |
| Show HN |
| Get comments |
💻 Code Execution (4)
Tool | Description |
| Sandboxed execution (Python/JS/Bash) |
| Python with common imports |
| Test code output |
📊 Data & Charts (11)
Tool | Description |
| Create DataFrame |
| Filter data |
| Aggregate/group data |
| Compute correlation |
| Detect outliers |
| Line chart |
| Bar chart |
| Pie chart |
| Scatter plot |
| Histogram |
| Ant Design spec |
🧠 Engineering Intelligence (17)
Tool | Description |
| Classify task type |
| Find relevant files |
| Get execution flow |
| Pinpoint bug causes |
| Generate implementation plan |
| Scaffold POC |
| Estimate blast radius |
| Select minimum tests |
| Build documentation |
| Plan docs updates |
| Save task state |
| Restore checkpoint |
| Create thinking session |
| Add reasoning step |
| Get summary |
| Structured analysis |
⚡ Optimization Tools (14)
Tool | Description |
| Token optimization stats |
| Minimal token fetch |
| Optimized extraction |
| Structured extraction |
| CSS selector extraction |
| Store output externally |
| Retrieve stored output |
| Search stored outputs |
| Session summary |
| Context mode stats |
| Lazy tool list |
| Load schemas on-demand |
| Search tools |
| Natural language search |
💾 Files & HuggingFace (10)
Tool | Description |
| Read file |
| Write file |
| List directory |
| Search files |
| Search models |
| Search datasets |
| Model info |
| Trending models |
| HTML → Markdown |
| URL → Markdown |
Architecture
┌─────────────────────────────────────────────────────────────────┐
│ Claude Code │
│ ┌───────────────────────────────────────────────────────────┐ │
│ │ MCP Servers (< 10 recommended) │ │
│ │ │ │
│ │ • mcp-server (83 tools) → Port 7710 │ │
│ │ • github (10 tools) → NPX │ │
│ │ • memory (5 tools) → NPX │ │
│ │ • sentry (5 tools) → NPX │ │
│ └───────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ Docker Services (Local) │
│ │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ │
│ │ MCP Server │ │ PostgreSQL │ │ Redis │ │
│ │ :7710 │ │ :7173 │ │ :7174 │ │
│ └──────────────┘ └──────────────┘ └──────────────┘ │
│ │
│ ┌──────────────┐ ┌──────────────┐ │
│ │ ChromaDB │ │ Firecrawl │ │
│ │ :8000 │ │ :7172 │ │
│ └──────────────┘ └──────────────┘ │
└─────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ Remote (Pi5) │
│ │
│ ┌──────────────┐ │
│ │ SearXNG │ │
│ │ :7171 │ │
│ │ (Search API) │ │
│ └──────────────┘ │
└─────────────────────────────────────────────────────────────────┘Token Flow
┌─────────────────────────────────────────────────────────────────┐
│ Before Optimization │
│ │
│ Tool Schemas: 90 tools × 150 tokens = 13,500 tokens │
│ Web Fetch: ~8,000 tokens per page │
│ Tool Outputs: Full JSON in context │
│ Total: ~200,000 tokens per session │
└─────────────────────────────────────────────────────────────────┘
↓
┌─────────────────────────────────────────────────────────────────┐
│ After Optimization │
│ │
│ Tool Schemas: 64 tools × 42 tokens = 2,700 tokens (-80%) │
│ Web Fetch: Quick fetch with token budget = ~300 tokens (-96%)│
│ Tool Outputs: External reference = ~50 tokens (-98%) │
│ Total: ~30,000 tokens per session (-85%) │
└─────────────────────────────────────────────────────────────────┘Essential MCP Servers
We recommend these additional MCP servers for maximum productivity:
Must-Have (⭐⭐⭐)
Server | Description | Setup |
GitHub | Repository, issues, PRs, commits |
|
Memory | Persistent knowledge across sessions |
|
Sentry | Error tracking and debugging |
|
Recommended (⭐⭐)
Server | Description | Setup |
Cloudflare | Workers, KV, R2, D1 |
|
Slack | Channel messaging |
|
PostgreSQL | Database queries |
|
Puppeteer | Browser automation |
|
Complete Configuration
{
"mcpServers": {
"mcp-server": {
"command": "docker",
"args": ["exec", "-i", "mcp-server", "python", "-c", "from mcp_server.server import run; run()"]
},
"github": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-github"]
},
"memory": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-memory"]
},
"sentry": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-sentry"]
},
"cloudflare": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-cloudflare"]
}
}
}Total tools: ~110 tools across 5 servers
Docker Services
Service | Port | Description |
MCP Server | 7710 | Main MCP protocol server (83 tools) |
SearXNG | 7171 | Self-hosted web search (Pi5) |
Firecrawl | 7172 | Advanced web scraping |
PostgreSQL | 7173 | Database for persistence |
Redis | 7174 | Cache and job queue |
ChromaDB | 8000 | Vector database for RAG |
Installation
Prerequisites
Docker & Docker Compose
Node.js (optional, for NPX MCPs)
Pi5 IP (for remote SearXNG, optional)
Steps
# 1. Clone
git clone https://github.com/simpletoolsindia/extra_skills_mcp_tools.git
cd extra_skills_mcp_tools
# 2. Run installer (follow prompts)
./install-claude-code.sh
# 3. Enter Pi5 IP when prompted (default: 192.168.0.2)
# Or press Enter to skip (SearXNG will be unavailable)
# 4. Restart Claude Code
claude
# 5. Verify
docker compose -f docker-compose.local.yml psManual Setup
# Start Docker services
docker compose -f docker-compose.local.yml up -d
# Add to ~/.claude/settings.json
cat >> ~/.claude/settings.json << 'EOF'
{
"mcpServers": {
"mcp-server": {
"command": "docker",
"args": ["exec", "-i", "mcp-server", "python", "-c", "from mcp_server.server import run; run()"]
}
}
}
EOFConfiguration
Environment Variables
# Pi5 (Remote) SearXNG
export SEARXNG_BASE_URL=https://your-pi5-ip:7171
# Local Ollama (optional)
export OLLAMA_BASE_URL=http://localhost:11434
# PostgreSQL
export POSTGRES_HOST=localhost
export POSTGRES_PORT=7173
export POSTGRES_DB=mcp_server
export POSTGRES_USER=mcp_user
export POSTGRES_PASSWORD=postgres
# Redis
export REDIS_HOST=localhost
export REDIS_PORT=7174
# Claude Code Optimization
export MAX_THINKING_TOKENS=10000
export CLAUDE_AUTOCOMPACT_PCT_OVERRIDE=50Pi5 (Remote) Setup
For remote SearXNG on Pi5:
# On Pi5
git clone https://github.com/simpletoolsindia/extra_skills_mcp_tools.git
cd extra_skills_mcp_tools
./start-remote.shThen enter Pi5 IP during installation.
Commands
# Start all services
docker compose -f docker-compose.local.yml up -d
# Stop all services
docker compose -f docker-compose.local.yml down
# View logs
docker compose -f docker-compose.local.yml logs -f mcp-server
# Rebuild after changes
docker compose -f docker-compose.local.yml up -d --build
# Quick restart
./start.sh # Start
./stop.sh # StopTesting
# Test token optimization
docker exec mcp-server python -c "
from src.mcp_server.server import _get_token_stats
import json
print(json.dumps(_get_token_stats(), indent=2))
"
# Test web search
docker exec mcp-server python -c "
from src.mcp_server.tools.searxng import search
print(search('MCP token optimization', limit=3))
"
# Test web fetch
docker exec mcp-server python -c "
from src.mcp_server.tools.web_fetch_optimized import quick_fetch
result = quick_fetch('https://example.com', max_tokens=500)
print(f'Title: {result[\"title\"]}')
print(f'Tokens: {result[\"tokens_used\"]}')
"
# Test MCP via network
echo '{"jsonrpc":"2.0","method":"tools/call","params":{"name":"get_token_stats","arguments":{}},"id":1}' | nc localhost 7710Troubleshooting
SearXNG returns 403
# Ensure Pi5 SearXNG has limiter disabled
ssh pi5 "docker exec searxng sed -i 's/limiter: true/limiter: false/' /etc/searxng/settings.yml && docker restart searxng"SSL Certificate Error
# Rebuild MCP server (includes ca-certificates)
docker compose -f docker-compose.local.yml up -d --build mcp-serverChromaDB Connection Error
# Check and restart
docker compose restart chromadbHigh Token Usage
Use
/costto monitorEnable only needed MCP servers
Use
quick_fetchinstead offetch_web_contentStore large outputs with
ctx_store_outputCompact at 50% with
/compact
Documentation
Document | Description |
| This file |
| Technical implementation details |
| Claude Code settings & strategies |
| Curated MCP server list |
Research Sources
Speakeasy Dynamic Toolsets — 91-97% reduction
Scott Spence Optimization — 60% reduction
Firecrawl Token Optimization — 97.9% HTML reduction
Mintlify Token Guide — 70% cost reduction
The New Stack - 10 Strategies — Architecture patterns
Context Mode MCP — 98% output reduction
Awesome MCP Servers — Curated server list
punkpeye/awesome-mcp-servers — Production-ready servers
win4r/Awesome-Claude-MCP-Servers — Claude-optimized
License
MIT License
Star History
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Latest Blog Posts
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
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/simpletoolsindia/extra_skills_mcp_tools'
If you have feedback or need assistance with the MCP directory API, please join our Discord server