MCP Server for Claude Desktop

The Protocol

What Is the Model Context Protocol?

The Model Context Protocol (MCP) is an open standard created by Anthropic that lets external tools connect directly to Claude Desktop. Instead of copy-pasting between windows or writing wrapper scripts, MCP gives Claude structured access to real capabilities -- file systems, APIs, databases, and custom tooling -- through a single stdio connection.

Think of MCP as a USB port for AI. Claude Desktop is the computer, and MCP servers are the peripherals. Each server exposes a set of tools that Claude can call natively during a conversation. No browser extensions. No middleware. Just a JSON config entry and a binary that speaks the protocol.

chAIrman is an MCP server built specifically for software development orchestration. When you connect it to Claude Desktop, Claude gains the ability to hire, manage, and coordinate an entire workforce of Claude Code agents -- turning a single chat window into a full project management system.

// claude_desktop_config.json { "mcpServers": { "chairman": { "command": "./chairman-macos-arm64", "args": [] } } } // That's it. Restart Claude Desktop. // 54 tools are now available.

54 MCP Tools

Everything You Need to Run a Virtual Dev Team

chAIrman organizes its 54 tools into logical groups. Each one is callable directly by Claude during your conversation.

T

Team Management

Hire agents, fire them, list your active team. Each agent gets a role, a model assignment (Sonnet or Opus), and a job description that serves as its permanent context. Agents persist across tasks until you dismiss them.

P

Pipeline Orchestration

Assign tasks with dependency chains using the depends_on parameter. Agents queue automatically until prerequisites finish. Cycle detection prevents deadlocks. Critical path calculation estimates completion time.

B

Backlog & Tickets

Create structured backlogs with milestones and tickets before any code is written. Each ticket maps to one agent and includes file ownership, success criteria, and off-limits files to prevent merge conflicts.

M

Memory & Continuity

Handoff documents, alumni archives, CEO notes, and project briefings ensure no context is lost between sessions. Rehire veteran agents who already know your codebase. Search across all memory with keyword queries.

$

Budget & Metrics

Set USD spending caps per project. Track cost per task, per agent, and per model. View efficiency scores, throughput rates, and cost trend analysis. Get optimization suggestions based on actual spend data.

H

Health & Webhooks

Monitor system memory, CPU load, and zombie processes. Detect blocked pipelines and high error rates. Subscribe external systems to events via HMAC-signed webhooks for task completion, budget warnings, and pipeline status.

Architecture

How the MCP Server Actually Works

When you type a request in Claude Desktop, Claude recognizes which chAIrman tools to call and invokes them over stdio. The MCP server receives the tool call, executes the logic (spawning agents, reading state, managing pipelines), and returns structured results that Claude uses to continue the conversation.

Agent processes are spawned as separate Claude Code instances, each running with full filesystem access in your project directory. Their output is parsed as stream-json, giving chAIrman real-time visibility into tool calls, file writes, costs, and completion status.

Server communicates over stdio -- no HTTP, no ports, no network exposure
State stored locally in ~/.claude-orchestrator/state.json with atomic writes
Dashboard runs on localhost:3456 with WebSocket real-time updates
Agent prompts assembled server-side with role, context files, skills, and messages
Git auto-commit after every successful task completion
Skills auto-discovered via TF-IDF matching from an 857-file library

# The flow: You (Claude Desktop) | | MCP tool calls over stdio v chAIrman (MCP Server) | | spawns Claude Code processes | manages state, memory, pipelines v Agents (Claude Code instances) | | autonomous work in project dir | stdout parsed as stream-json v Results -> handoffs, git commits, metrics, alumni archive

2-Minute Setup

Install the MCP Server and Start Building

chAIrman ships as a native binary. No Node.js runtime, no npm install, no Docker. Download, configure, restart Claude Desktop.

54

MCP Tools

857

Skill Files

10

Agent Templates

2 min

Setup Time

# Step 1: Download the binary curl -L https://mrchairman.ai/download \ -o chairman-macos-arm64 chmod +x chairman-macos-arm64 # Step 2: Add to Claude Desktop config # Open: ~/Library/Application Support/ # Claude/claude_desktop_config.json { "mcpServers": { "chairman": { "command": "/path/to/chairman-macos-arm64" } } } # Step 3: Restart Claude Desktop # Done. Ask Claude to create a project.

Three Steps. Zero Dependencies.

chAIrman compiles to a standalone binary for macOS (ARM64 and x64) and Linux (x64). There is no runtime to install, no package manager to configure, and no containerization required. The binary includes everything: the MCP server, the dashboard, the skills engine, and the agent orchestrator.

Once you add the binary path to your Claude Desktop configuration file and restart, all 54 tools appear automatically in Claude's tool palette. You can verify the connection by asking Claude to run list_projects or health_check.

The dashboard launches automatically at localhost:3456, giving you a visual kanban board, live agent terminals, pipeline visualization, budget tracking, and git history -- all updating in real time via WebSocket.

Common Questions

MCP Server FAQ

What is an MCP server and why do I need one?

An MCP (Model Context Protocol) server is a program that exposes tools to Claude Desktop over a standardized stdio protocol. Without an MCP server, Claude can only chat. With one, Claude can call external functions -- read files, query databases, manage processes, or in chAIrman's case, orchestrate an entire team of AI agents. You need an MCP server whenever you want Claude to do something beyond text generation.

How is chAIrman different from other MCP servers?

Most MCP servers expose a handful of tools for a specific integration (Slack, GitHub, a database). chAIrman exposes 54 tools that turn Claude into a project manager. It can hire agents, assign tasks with dependency chains, track budgets, auto-retry failures, maintain memory across sessions, auto-commit to git, and monitor system health. It is a complete orchestration layer, not a single-purpose connector.

Does chAIrman work with Claude Code or only Claude Desktop?

chAIrman connects to Claude Desktop as an MCP server. It then spawns Claude Code agents as child processes to do the actual work. So you interact with chAIrman through Claude Desktop, but the agents it hires are Claude Code instances running autonomously in your project directory. You get the best of both: Claude Desktop's conversation interface and Claude Code's filesystem capabilities.

Is my code sent to external servers?

Agent prompts are assembled via the chAIrman API (for skill injection and prompt optimization), but your actual project files stay local. Agents run as local Claude Code processes with filesystem access to your project directory. State, handoffs, metrics, and alumni archives are stored in ~/.claude-orchestrator/ on your machine. The dashboard runs on localhost only.

What platforms does the MCP server support?

chAIrman ships as pre-built binaries for macOS ARM64 (Apple Silicon), macOS x64 (Intel), and Linux x64. Windows support is not currently available. The binary is self-contained with no external dependencies -- you do not need Node.js, Python, or Docker installed.

How much does it cost?

chAIrman starts at $19.99/month for the Pro tier (2 concurrent agents, 5 projects) and $99.99/month for Unlimited (no caps on agents or projects). You also pay standard Anthropic API costs for the Claude Code agents chAIrman spawns. chAIrman includes per-project budget controls so you can set USD spending caps and get alerts at 80% utilization.

The MCP Server That Turns Claude Into a CEO