Abstract

Abstract MCP Server acts as a smart proxy that caches large tool responses from MCP servers, keeping your conversation context lean while still giving you access to full results via compact links and optional file uploads. It supports direct tool calls and flexible storage, so you can scale your MCP workflow without bloating prompts.

How to use

You integrate Abstract into your MCP client workflow to route tool calls through a white-listed proxy that caches large responses and forwards requests to upstream MCP servers. Use Abstract when you want to minimize context window usage while still retaining easy access to full results via resource links, with support for file uploads and format conversions.

Configure your MCP client to include an Abstract server alongside your existing MCP servers. Abstract will read your client config to discover which upstream servers to proxy and will enforce storage directories you allow. When you perform tool calls, Abstract may cache outputs to files and return compact links to those resources, or pass through small responses directly when appropriate.

How to install

Prerequisites you need before installation: Node.js and npm. You will also need a directory you’re willing to designate as storage for cached responses.

# Install Abstract MCP globally
npm install -g abstract-mcp

# Or build from source if you prefer
# git clone https://example.com/abstract-mcp.git
# cd abstract-mcp
# npm install
# npm run build

Configuration and usage patterns

Add Abstract as a named MCP server in your client configuration. You provide the storage directories Abstract may write to, along with environment overrides that point Abstract to your client config and upstream servers.

{
  "mcpServers": {    
    "abstract": {
      "command": "abstract-mcp",
      "args": [
        "/path/to/allowed/storage/dir1",
        "/path/to/allowed/storage/dir2"
      ],
      "env": {
        "APP_CONFIG_PATH": "/path/to/claude/desktop/config.json",
        "ABSTRACT_PROXY_SERVERS": "server1,server2"
      }
    }
  }
}

Example with an upstream MCP server

You can pair Abstract with an upstream MCP server such as Tavily MCP. The following configuration shows how Abstract delegates to an upstream server when caching is involved.

{
  "mcpServers": {    
    "abstract": {
      "command": "abstract-mcp",
      "args": [
        "/path/to/allowed/storage/dir1",
        "/path/to/allowed/storage/dir2"
      ],
      "env": {
        "APP_CONFIG_PATH": "/path/to/claude/desktop/config.json",
        "ABSTRACT_PROXY_SERVERS": "tavily-mcp"
      }
    },
    "tavily-mcp": {
      "command": "npx",
      "args": ["-y", "tavily-mcp@latest"],
      "env": {
        "TAVILY_API_KEY": "your-tavily-api-key"
      }
    }
  }
}

Available tools

call_tool_and_store

Caches large responses to files and returns compact resource links to keep the conversation context lean.

call_tool

Direct tool calls in the current context without caching.

call_tool_with_file_content

Uploads file content to tools to enable bulk processing while avoiding context bloat.

list_available_tools

Discovers upstream tools available on configured MCP servers.

list_tool_details

Fetches complete parameter schemas for a specific tool.

list_allowed_directories

Lists directories Abstract is allowed to use for storage.