slack-cli-mcp

"We've been using MCP wrong. Loading all tool definitions upfront consumed ~150,000 tokens before any user request." — Anthropic Engineering Blog

A Docker-wrapped Slack MCP server exposed as CLI, achieving 98.7% token reduction for LLM agents.

Quick Start

# 1. Build
./slack-cli-mcp docker build

# 2. Start + Login
./slack-cli-mcp docker start
./slack-cli-mcp docker creds-login

# 3. Use
./slack-cli-mcp list-channels

Why CLI Over MCP?

| Approach | Token Usage | LLM Accuracy | |----------|-------------|--------------| | Traditional MCP | ~150,000 tokens | Degrades with >2-3 servers | | CLI Interface | ~2,000 tokens | Consistent |

Benefits:

Token efficiency: Only load what you need
Scriptable: LLMs compose commands, write wrappers
Composable: Pipes, redirects, standard shell tools

For the full rationale, see MCP Considered Suboptimal — a technical knowledge base with industry analysis and Anthropic engineering evidence.

Architecture

User/Agent → slack-cli-mcp → Docker Container → Slack API
                               ├── slack-mcp-server
                               └── mcptools

See docs/architecture.md for details.

Documentation

Tool Reference - Full command documentation
Architecture - System design
Design Principles - Why we built it this way

Components

korotovsky/slack-mcp-server - Slack MCP server
f/mcptools - MCP-to-CLI bridge

Future Work

See FUTURE_WORK.md for roadmap including:

Slack server mock for testing
Network isolation
Official Slack MCP integration (when available)

Note: Slack is developing an official MCP server. When available, we'll evaluate integration. File an issue if you'd like to help!

Research

This project is informed by:

MCP Considered Suboptimal - Technical knowledge base on CLI-over-MCP efficiency
- THESIS.md - Core arguments and triadic failure model
- CLI-SDK-over-context-bloat.md - Token reduction patterns
Anthropic's 98.7% cost reduction
Linearis: CLI for AI agents
The Rise of Industrial Software

License

MIT

MCP Servers