Introduction to SLOP MCP
Install 17 MCPs. Claude won't even notice.
SLOP MCP is a Model Context Protocol (MCP) orchestration layer that lets you connect unlimited MCP servers to Claude without destroying your context window.
The Problem
Traditional MCP setups have a fatal flaw: every tool from every MCP floods your context. Install 5 MCPs with 20 tools each, and you've just burned 100+ tool schemas before asking your first question.
Without SLOP MCP:
┌─────────────────────────────────────────────┐
│ Context Window (200k tokens) │
├─────────────────────────────────────────────┤
│ 🔧 Tool schemas from MCP 1 (20 tools) │
│ 🔧 Tool schemas from MCP 2 (15 tools) │
│ 🔧 Tool schemas from MCP 3 (30 tools) │
│ 🔧 Tool schemas from MCP 4 (25 tools) │
│ 🔧 Tool schemas from MCP 5 (10 tools) │
│ ─────────────────────────────────────────── │
│ 📝 Your actual conversation... (what's left)│
└─────────────────────────────────────────────┘
As described in Anthropic's Code Execution with MCP article, this context overload is a fundamental challenge. SLOP MCP addresses it with progressive tool discovery and in-environment script execution via the SLOP language.
The Solution
SLOP MCP exposes just 8 meta-tools regardless of how many MCPs you connect. Tool schemas are loaded on-demand when you search for them.
With SLOP MCP:
┌─────────────────────────────────────────────┐
│ Context Window (200k tokens) │
├─────────────────────────────────────────────┤
│ 🔧 search_tools 🔧 execute_tool │
│ 🔧 manage_mcps 🔧 auth_mcp │
│ 🔧 get_metadata 🔧 run_slop │
│ 🔧 slop_reference 🔧 slop_help │
│ ─────────────────────────────────────────── │
│ 📝 Your entire conversation fits here! │
│ │
│ (Tools loaded only when you need them) │
└─────────────────────────────────────────────┘
Key Features
🚀 Unlimited MCPs, Constant Context
Connect 17 MCPs or 170. Your context overhead stays the same: 8 tool schemas.
🔍 On-Demand Tool Discovery
Use search_tools to find what you need. Only matching tools get loaded into context.
# Search across all MCPs
search_tools query="file upload"
# Found 3 tools from 2 MCPs
# Only these 3 schemas hit your context!
⚡ Skills = Zero Overhead
Create skills that call MCPs directly. The skill invocation is one line; SLOP handles the orchestration behind the scenes.
# In Claude Code
/calculate-tax income=50000 state=CA
# Skill calls math-mcp and tax-mcp internally
# No tool schemas in context at all!
🔐 OAuth That Actually Works
Authenticate to MCPs like Figma, Linear, or Dart with browser-based OAuth. Connection auto-reconnects after auth.
slop-mcp mcp auth login figma
# Browser opens, you authenticate
# MCP reconnects automatically with new credentials
📦 Flexible Configuration
Define MCPs at project, user, or local scope using KDL config files:
mcp "math-mcp" {
transport "stdio"
command "npx"
args "@anthropic/math-mcp"
}
mcp "figma" {
transport "streamable"
url "https://mcp.figma.com/mcp"
}
Quick Example
Here's SLOP MCP in action with Claude Code:
User: Calculate 15% tip on $47.50
Claude: Let me search for calculation tools.
> search_tools query="calculate"
Found: math-mcp:calculate, math-mcp:evaluate
> execute_tool mcp_name="math-mcp" tool_name="calculate"
parameters={"expression": "47.50 * 0.15"}
Result: 7.125
The 15% tip on $47.50 is $7.13.
Only the calculate tool schema was loaded - not the 50+ other tools from your connected MCPs.