infographic compare-binary-horizontal-simple-fold
data
  title RLM Architecture Comparison
  items
    - label Traditional Approach
      desc Server calls external LLM API
      icon mdi:server-network
    - label This Server (v2.0)
      desc Client LLM does all reasoning
      icon mdi:brain

infographic sequence-steps-simple
data
  title RLM Processing Flow
  items
    - label 1. Load
      desc Load long context into server
    - label 2. Analyze
      desc Get structure and statistics
    - label 3. Decompose
      desc Split into manageable chunks
    - label 4. Process
      desc LLM reasons over chunks
    - label 5. Aggregate
      desc Combine into final answer

# Clone or navigate to project
cd rlm-mcp-server

# Install dependencies
npm install

# Build
npm run build

# Run
npm start

{
  "mcpServers": {
    "rlm": {
      "command": "node",
      "args": ["C:\\path\\to\\rlm-mcp-server\\dist\\index.js"]
    }
  }
}

{
  "mcpServers": {
    "rlm": {
      "command": "node",
      "args": ["/path/to/rlm-mcp-server/dist/index.js"]
    }
  }
}

{
  "rlm-mcp-server": {
    "command": "node",
    "args": ["/path/to/rlm-mcp-server/dist/index.js"]
  }
}

// Output
print(...args)                    // Print to output

// Context
getContext(id)                    // Get full content
getContextMetadata(id)            // Get metadata

// String Operations
len(str)                          // Length
slice(str, start, end)            // Substring
split(str, sep)                   // Split to array
join(arr, sep)                    // Join to string
trim(str), lower(str), upper(str) // String transforms

// Regex
search(pattern, text, flags)      // Find matches
findAll(pattern, text)            // All matches with index
replace(text, pattern, repl)      // Replace

// Array
range(start, end, step)           // Generate range
map(arr, fn)                      // Transform
filter(arr, fn)                   // Filter
reduce(arr, fn, init)             // Reduce
sort(arr, fn)                     // Sort (copy)
unique(arr)                       // Remove duplicates
chunk(arr, size)                  // Split array

// Variables
setVar(name, value)               // Store
getVar(name)                      // Retrieve
listVars()                        // List all

// Answer
setAnswer(content, ready)         // Set answer
getAnswer()                       // Get answer state

// JSON
JSON.parse(str)                   // Parse
JSON.stringify(obj, indent)       // Stringify

1. Load the document:
   rlm_load_context(context="...", context_id="doc")

2. Analyze structure:
   rlm_get_context_info(context_id="doc")
   → Returns: 500,000 chars, markdown, 12,000 lines

3. Get strategy suggestion:
   rlm_suggest_strategy(context_id="doc")
   → Returns: by_sections (markdown content)

4. Decompose:
   rlm_decompose_context(context_id="doc", strategy="by_sections")
   → Returns: 45 chunks (sections)

5. Search for relevant sections:
   rlm_search_context(context_id="doc", pattern="climate change")
   → Returns: Matches in chunks 3, 7, 12, 23

6. Get those chunks:
   rlm_get_chunks(chunk_indices=[3, 7, 12, 23])
   → Returns: Content of those sections

7. Process each chunk (LLM reasoning)
   Build understanding from each section...

8. Save intermediate results:
   rlm_set_variable(name="findings", value=[...])

9. Aggregate into final answer:
   rlm_set_answer(content="Based on analysis...", ready=true)

┌─────────────────────────────────────────────────────────┐
│                     MCP Client                          │
│  ┌─────────────────────────────────────────────────┐   │
│  │                Your LLM                          │   │
│  │  (Claude, GPT, Llama, Gemini, etc.)             │   │
│  │                                                  │   │
│  │  Performs all reasoning and recursive calls     │   │
│  └─────────────────────────────────────────────────┘   │
│                         │                               │
│                    MCP Protocol                         │
│                         │                               │
└─────────────────────────┼───────────────────────────────┘
                          │
┌─────────────────────────┼───────────────────────────────┐
│              RLM MCP Server (this)                      │
│                         │                               │
│  ┌──────────────────────┴──────────────────────────┐   │
│  │              Tools Layer                         │   │
│  │  load, read, decompose, search, execute, etc.   │   │
│  └─────────────────────────────────────────────────┘   │
│                         │                               │
│  ┌──────────────────────┴──────────────────────────┐   │
│  │            Services Layer                        │   │
│  │  ┌─────────────┐  ┌────────────────────────┐   │   │
│  │  │  Session    │  │  Context Processor     │   │   │
│  │  │  Manager    │  │  (decompose, search)   │   │   │
│  │  └─────────────┘  └────────────────────────┘   │   │
│  └─────────────────────────────────────────────────┘   │
│                                                         │
│  No external dependencies - pure JavaScript             │
└─────────────────────────────────────────────────────────┘

node dist/index.js

node dist/index.js --http --port=3000