mcp

Arkestone MCP Servers

A suite of Model Context Protocol (MCP) servers for GitHub Copilot and other AI coding assistants. Each server dynamically serves a different type of context — custom instructions, skills, prompts, ADRs, persistent memory, or knowledge graphs — from local directories and GitHub repositories.

Built with the Go MCP SDK. Browse the MCP Registry for the full ecosystem of MCP servers.

Table of Contents

• Available Servers
• Installation
  • go install
  • Docker
  • Pre-built Binaries
  • Build from Source
• Quick Start
• Transport Mechanisms
• MCP Client Configuration
  • VS Code
  • Claude Desktop
  • Cursor
  • Windsurf
  • Claude Code
  • JetBrains
  • Zed
  • Docker Compose (all servers)
• LLM Optimization
• GitHub Authentication
• Network & Proxy
• Architecture
• Shared Packages
• Development
• Contributing

Available Servers

Each server has its own README and CHANGELOG.

Installation

go install

The fastest way to install. Requires Go 1.24+.

# Install the latest version
go install github.com/Arkestone/mcp/servers/mcp-instructions/cmd/mcp-instructions@latest
go install github.com/Arkestone/mcp/servers/mcp-skills/cmd/mcp-skills@latest
go install github.com/Arkestone/mcp/servers/mcp-prompts/cmd/mcp-prompts@latest
go install github.com/Arkestone/mcp/servers/mcp-adr/cmd/mcp-adr@latest
go install github.com/Arkestone/mcp/servers/mcp-memory/cmd/mcp-memory@latest
go install github.com/Arkestone/mcp/servers/mcp-graph/cmd/mcp-graph@latest

# Install a pinned version
go install github.com/Arkestone/mcp/servers/mcp-instructions/cmd/mcp-instructions@v0.0.1

Docker

Pre-built images are published to the GitHub Container Registry after each release:

# Pull the latest version
docker pull ghcr.io/arkestone/mcp-instructions:latest
docker pull ghcr.io/arkestone/mcp-skills:latest
docker pull ghcr.io/arkestone/mcp-prompts:latest
docker pull ghcr.io/arkestone/mcp-adr:latest
docker pull ghcr.io/arkestone/mcp-memory:latest
docker pull ghcr.io/arkestone/mcp-graph:latest

# Pull a specific version
docker pull ghcr.io/arkestone/mcp-instructions:v0.0.1

Pre-built Binaries

Download pre-built binaries for Linux, macOS, and Windows from GitHub Releases:

# Example: Linux amd64
curl -L https://github.com/Arkestone/mcp/releases/latest/download/mcp-instructions_linux_amd64.tar.gz | tar xz
mv mcp-instructions /usr/local/bin/

Build from Source

git clone https://github.com/Arkestone/mcp.git
cd mcp
make build    # → ./bin/mcp-instructions, ./bin/mcp-skills, ...

Quick Start

# stdio — used by VS Code, Claude Desktop, Cursor, etc.
mcp-instructions -dirs /path/to/repo

# HTTP — for shared team deployments
mcp-instructions -transport http -addr :8080 -repos github/awesome-copilot

# With LLM optimization
mcp-instructions -transport http -llm.enabled -llm.endpoint https://api.openai.com/v1

See each server's README for the full configuration reference.

Transport Mechanisms

All servers default to stdio. Pass -transport http to switch to HTTP.

Quick Install in VS Code

Click to install any server directly into VS Code or VS Code Insiders. stdio requires the binary installed locally (go install); HTTP connects to a running local server; Docker runs the server in a container (no local install needed). File-scanning servers use ${workspaceFolder} — open a folder first.

Server	Transport	VS Code	VS Code Insiders
mcp-instructions	stdio
	HTTP
	Docker
mcp-skills	stdio
	HTTP
	Docker
mcp-prompts	stdio
	HTTP
	Docker
mcp-adr	stdio
	HTTP
	Docker
mcp-memory	stdio
	HTTP
	Docker
mcp-graph	stdio
	HTTP
	Docker

MCP Client Configuration

VS Code

Quick install: Use the badges in Quick Install in VS Code above, or open the Command Palette (Ctrl+Shift+P / Cmd+Shift+P) and run MCP: Open User Configuration to add a server to your global mcp.json. See the VS Code MCP documentation for full details.

.vscode/mcp.json (project-level)

{
  "servers": {
    "instructions": { "command": "mcp-instructions", "args": ["-dirs", "${workspaceFolder}"] },
    "skills":       { "command": "mcp-skills",       "args": ["-dirs", "${workspaceFolder}"] },
    "prompts":      { "command": "mcp-prompts",      "args": ["-dirs", "${workspaceFolder}"] },
    "adrs":         { "command": "mcp-adr",          "args": ["-dirs", "${workspaceFolder}"] },
    "memory":       { "command": "mcp-memory" },
    "graph":        { "command": "mcp-graph" }
  }
}

Global user settings (settings.json)

{
  "github.copilot.chat.mcp.servers": {
    "instructions": {
      "command": "mcp-instructions",
      "args": ["-dirs", "/path/to/repo"]
    }
  }
}

Claude Desktop

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "instructions": {
      "command": "mcp-instructions",
      "args": ["-dirs", "/path/to/repo"]
    },
    "skills": {
      "command": "mcp-skills",
      "args": ["-dirs", "/path/to/repo"]
    },
    "memory": {
      "command": "mcp-memory",
      "env": { "MEMORY_DIR": "~/.local/share/mcp-memory" }
    },
    "graph": {
      "command": "mcp-graph"
    }
  }
}

Cursor

.cursor/mcp.json (project-level)

{
  "mcpServers": {
    "instructions": {
      "command": "mcp-instructions",
      "args": ["-dirs", "."]
    },
    "skills": {
      "command": "mcp-skills",
      "args": ["-dirs", "."]
    },
    "prompts": {
      "command": "mcp-prompts",
      "args": ["-dirs", "."]
    },
    "memory": {
      "command": "mcp-memory"
    }
  }
}

Global (~/.cursor/mcp.json)

{
  "mcpServers": {
    "memory": {
      "command": "mcp-memory",
      "env": { "MEMORY_DIR": "~/.local/share/mcp-memory" }
    }
  }
}

Windsurf

macOS/Linux: ~/.codeium/windsurf/mcp_config.json Windows: %USERPROFILE%\.codeium\windsurf\mcp_config.json

{
  "mcpServers": {
    "instructions": {
      "command": "mcp-instructions",
      "args": ["-dirs", "/path/to/repo"]
    },
    "memory": {
      "command": "mcp-memory",
      "env": { "MEMORY_DIR": "~/.local/share/mcp-memory" }
    }
  }
}

Claude Code

~/.mcp.json (global) or .mcp.json (project-level)

{
  "mcpServers": {
    "instructions": {
      "command": "mcp-instructions",
      "args": ["-dirs", "."]
    },
    "skills": {
      "command": "mcp-skills",
      "args": ["-dirs", "."]
    },
    "memory": {
      "command": "mcp-memory"
    },
    "graph": {
      "command": "mcp-graph"
    }
  }
}

JetBrains

In JetBrains IDEs (IntelliJ IDEA, GoLand, PyCharm, etc.), go to Settings → Tools → AI Assistant → Model Context Protocol (MCP):

{
  "mcpServers": {
    "instructions": {
      "command": "mcp-instructions",
      "args": ["-dirs", "$PROJECT_DIR$"]
    },
    "memory": {
      "command": "mcp-memory"
    }
  }
}

Zed

Add to ~/.config/zed/settings.json:

{
  "context_servers": {
    "mcp-instructions": {
      "command": {
        "path": "mcp-instructions",
        "args": ["-dirs", "/path/to/repo"]
      }
    },
    "mcp-memory": {
      "command": {
        "path": "mcp-memory",
        "args": []
      }
    }
  }
}

Docker Compose — all servers

services:
  mcp-instructions:
    image: ghcr.io/arkestone/mcp-instructions:latest
    ports: ["8080:8080"]
    environment:
      INSTRUCTIONS_TRANSPORT: http
      INSTRUCTIONS_DIRS: /data
    volumes: ["./instructions:/data:ro"]

  mcp-skills:
    image: ghcr.io/arkestone/mcp-skills:latest
    ports: ["8081:8081"]
    environment:
      SKILLS_TRANSPORT: http
      SKILLS_DIRS: /data
    volumes: ["./skills:/data:ro"]

  mcp-prompts:
    image: ghcr.io/arkestone/mcp-prompts:latest
    ports: ["8082:8082"]
    environment:
      PROMPTS_TRANSPORT: http
      PROMPTS_SOURCES_DIRS: /data
    volumes: ["./prompts:/data:ro"]

  mcp-adr:
    image: ghcr.io/arkestone/mcp-adr:latest
    ports: ["8083:8083"]
    environment:
      ADR_TRANSPORT: http
      ADR_DIRS: /data
    volumes: ["./docs:/data:ro"]

  mcp-memory:
    image: ghcr.io/arkestone/mcp-memory:latest
    ports: ["8084:8084"]
    environment:
      MEMORY_TRANSPORT: http
      MEMORY_DIR: /data
    volumes: ["memory-data:/data"]

  mcp-graph:
    image: ghcr.io/arkestone/mcp-graph:latest
    ports: ["8085:8085"]
    environment:
      GRAPH_TRANSPORT: http
      GRAPH_DATA_FILE: /data/graph.json
    volumes: ["graph-data:/data"]

volumes:
  memory-data:
  graph-data:

Then connect your client over HTTP:

{
  "mcpServers": {
    "instructions": {
      "type": "http",
      "url": "http://localhost:8080/mcp"
    }
  }
}

LLM Optimization

All content servers (instructions, skills, prompts, ADRs) optionally consolidate multiple sources using an OpenAI-compatible LLM endpoint. This deduplicates and logically organizes content from multiple repositories.

mcp-instructions \
  -dirs . \
  -llm.enabled \
  -llm.endpoint https://api.openai.com/v1 \
  -llm.api-key $LLM_API_KEY \
  -llm.model gpt-4o-mini

Supported providers: OpenAI, Azure OpenAI, Ollama, LM Studio, and any OpenAI-compatible endpoint.

See config.llm.example.yaml for a complete configuration example.

GitHub Authentication (Optional)

A GitHub token is optional. Public repositories work without authentication. For private repositories, provide a token (highest priority first):

Network & Proxy

All servers work on-premise, in private/public cloud, with direct internet or through HTTP/HTTPS proxies. See the Network & Proxy Guide for firewall rules, proxy configuration, and custom CA certificates.

Architecture

.
├── servers/
│   ├── mcp-instructions/   # custom instructions server  (:8080)
│   ├── mcp-skills/         # skills server               (:8081)
│   ├── mcp-prompts/        # prompt files server         (:8082)
│   ├── mcp-adr/            # ADR server                  (:8083)
│   ├── mcp-memory/         # persistent memory server    (:8084)
│   └── mcp-graph/          # knowledge graph server      (:8085)
├── pkg/
│   ├── config/             # shared configuration loading (YAML → env → flags)
│   ├── github/             # GitHub Contents API client
│   ├── httputil/           # proxy, TLS, header propagation
│   ├── optimizer/          # shared LLM optimization layer (OpenAI-compatible)
│   ├── server/             # MCP server bootstrap helpers
│   └── syncer/             # background repo sync
├── docs/
│   └── network.md          # network / proxy / firewall guide
├── examples/               # client configuration examples
└── AGENTS.md               # AI coding assistant guide

Each content server follows the same layered design:

1. Config — YAML → environment variables → CLI flags (each layer overrides the previous)
2. Loader / Scanner — discovers content from local directories and GitHub repositories
3. Optimizer — optional LLM-based consolidation via pkg/optimizer
4. MCP Server — exposes content as Resources, Prompts, and Tools over stdio or Streamable HTTP

Shared Packages

Development

make build              # build all servers into ./bin/
make test               # run unit tests
make test-integration   # run integration tests (requires LLM_ENDPOINT)
make docker             # build Docker images for all servers
make lint               # run golangci-lint
make cover              # generate coverage report

Contributing

Contributions are welcome! Please see CONTRIBUTING.md for guidelines.

License

This project is licensed under the MIT License — see the LICENSE file for details.