Massive Context MCP

You can run a Massive Context MCP server that loads large contexts externally, chunks them, runs sub-queries, and aggregates results for efficient analysis. It lets you keep massive data out of prompts while still enabling flexible, parallel querying and local or cloud inference as needed.

How to use

You will connect a client to the Massive Context MCP server and drive analysis through the provided tools. Load a large document as context, inspect its structure, and break it into manageable pieces. Then perform sub-queries on chunks in parallel, store the results, and retrieve an aggregated answer. The flow supports free local inference via Ollama when available, or cloud inference as a fallback.

How to install

Prerequisites you need before installing the server:

• Python and the uvx runtime must be available on your system.

Step 1: Install the MCP package

uvx massive-context-mcp

Or install with Python's package manager

pip install massive-context-mcp

Optional extras to enable security filters or Claude integration

pip install massive-context-mcp[firewall]
pip install massive-context-mcp[claude]

Step 2: Wire the MCP to Claude Code or Claude Desktop (example shown)

{
  "mcpServers": {
    "massive-context": {
      "command": "uvx",
      "args": ["massive-context-mcp"],
      "env": {
        "RLM_DATA_DIR": "~/.rlm-data",
        "OLLAMA_URL": "http://localhost:11434"
      }
    }
  }
}

Additional content

The MCP is designed around loading context as an external variable, inspecting its structure, chunking strategically (by lines, chars, or paragraphs), and performing recursive sub-queries on chunks. Results are aggregated to produce final synthesized output. This approach enables handling of contexts with tens of millions of tokens while keeping prompts compact and targeted.

Available tools

rlm_system_check

Check system requirements and readiness for running the MCP server.

rlm_setup_ollama

Install and configure Ollama via Homebrew, start as a service, and pull the required model.

rlm_ollama_status

Check if Ollama is running and what models are available for local inference.

rlm_load_context

Load a large document or dataset as an external context without embedding it into prompts.

rlm_inspect_context

Inspect the structure of a loaded context without loading content into the prompt.

rlm_chunk_context

Chunk a loaded context by lines, characters, or paragraphs for parallel processing.

rlm_sub_query

Run a sub-query on a specific context chunk, with optional provider selection.

rlm_sub_query_batch

Process multiple chunks in parallel to gather results quickly.

rlm_store_result

Store the result of a sub-query for later aggregation.

rlm_get_results

Retrieve stored results for aggregation or final synthesis.

rlm_exec

Execute arbitrary Python code against a loaded context in a sandboxed subprocess.

rlm_auto_analyze

Automatically analyze content, determine type, choose chunking strategy, and run parallel sub-queries.

rlm_filter_context

Filter loaded context with a regex to keep or remove specific lines.

rlm_ollama_status

Alias usage for checking Ollama status via the MCP tooling.

rlm_setup_ollama_direct

Direct download method for Ollama on headless or locked-down machines.

rlm_firewall_status

Check the status and configuration of the code firewall integration.