Slipsnisse

Slipsnisse (noun, Swedish): Lit. "Tie-Gnome". A slightly derogatory term for a corporate drone or "suit" who focuses on superficial process and career climbing rather than deep substance.

Slipsnisse is a hierarchical Model Context Protocol (MCP) server that acts as a "middle manager" subagent. It offloads verbose, context-heavy tasks from expensive orchestrator agents to faster, cheaper models.

How it works

Slipsnisse creates a semantic buffer:

1. The orchestrator (e.g., Claude Opus 4.5) sees a clean "Expert Tool".
2. Slipsnisse coordinates downstream MCPs (filesystem, docs, search) and cheaper "Flash" models (e.g., Gemini 3 Flash) to execute complex workflows.
3. The Result is synthesised and returned to the orchestrator, keeping its context window clean.

Architecture

graph LR
    Host[Orchestrator Agent] <-->|MCP| Slipsnisse[Slipsnisse Server]
    
    subgraph "Slipsnisse Runtime"
        Config[JSON Config]
        Manager[Client Manager]
        Loop[Subagent Loop]
    end
    
    Slipsnisse <--> Loop
    Loop <-->|Tool Calls| Manager
    
    Manager <-->|MCP| Downstream1[Filesystem]
    Manager <-->|MCP| Downstream2[Docs]

Loading

Features

• Hierarchical Delegation: Decouple expensive reasoning from verbose tool usage.
• Dynamic Tooling: Composite tools are defined in a simple JSON configuration.
• Vercel AI SDK Integration: Supports Google, OpenAI, Anthropic, and other providers.
• Environment Variable Substitution: Securely manage secrets in your config using ${VAR_NAME} syntax.
• Tool Namespacing: Automatically handles collisions between multiple downstream MCP servers.
• Low Overhead: Lightweight Node.js runtime using pnpm and tsx.

Getting Started

Installation

pnpm install

Usage

Launch Slipsnisse with a path to your configuration file:

npx slipsnisse --config ./config.json --log-level info

Using Additional Providers

Slipsnisse comes with core providers pre-installed. To use additional Vercel AI SDK providers (like @ai-sdk/azure or @ai-sdk/mistral) when running via npx, inject them using the -p flag:

npx -p slipsnisse -p @ai-sdk/mistral -p @ai-sdk/azure slipsnisse --config ./config.json

Configuration

Slipsnisse is driven by a JSON configuration file. You can find example configurations in the examples/ directory.

Environment Variables

You can use ${VAR_NAME} syntax anywhere in the configuration values (args, env, apiKey, etc.) to substitute parameters from the process environment.

Example Config

{
  "providers": {
    "google": {
      "provider": "google",
      "apiKey": "${GEMINI_API_KEY}"
    }
  },
  "mcps": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/src"]
    }
  },
  "tools": [
    {
      "name": "research_files",
      "description": "Analyses project files to answer specific questions.",
      "internal_tools": {
        "filesystem": ["read_file", "list_directory"]
      },
      "provider": "google",
      "model": "gemini-3-flash-exp",
      "temperature": 0.3,
      "system_prompt": "You are a senior engineer. Use the filesystem to research the codebase."
    }
  ]
}

CLI Arguments

• --config <path>: Path to JSON config (required).
• --log-level <level>: debug, info, warn, error (default: info).
• --log-pretty: Enable human-readable logging.

Security

Slipsnisse executes binaries defined in the mcps configuration. It should only be run in trusted environments. Tool whitelisting ensures subagents only access explicitly granted capabilities.

Licence

WIP, TBC