MCP Integration for AI Agents

ChangeSharp provides a built-in Model Context Protocol (MCP) server, allowing AI agents (like Claude, Cursor, or GitHub Copilot) to interact directly with your changelog and release workflow.

Features

The MCP server exposes the following tools:

• get_status: Get the count of pending fragments, the current version, and the next calculated version.
• create_fragment: Create a new change fragment with a message and a category. Mirrors the CLI new gate: optional allowMajor bypasses the SemverPolicy.MaxImpact cap when the category would force a Major bump.
• validate_fragments: Ensure all pending fragments follow the correct format. Optional apiMinLevel (patch | minor | major) mirrors the CLI --api-min-level gate: fails if the fragments' declared impact is below this level.
• perform_release: Execute a release, aggregate changes into the changelog, and bump project versions. Mirrors the CLI safety gates: optional allowMajor (bypass SemverPolicy.MaxImpact cap) and optional apiMinLevel.

The safety gates exposed by the CLI (--api-min-level, --allow-major / SemverPolicy.MaxImpact) behave identically through the MCP tools.

Security Warning: In enterprise environments, AI agents should NOT be allowed to perform a release without human approval. It is highly recommended to use the --dry-run flag or implement a mandatory approval gate in your CI/CD pipeline before the final release is pushed.

Security Configuration

The MCP server enforces a two-layer security model for the perform_release tool:

Setting	Default	Description
Security.AllowAgentRelease	true	Set to false in changesharp.json to block all MCP agent releases.
Security.RequireApproval	false	Set to true in changesharp.json to require CHANGESHARP_ALLOW_UNSAFE_RELEASE=true env var.
CHANGESHARP_ALLOW_UNSAFE_RELEASE	—	Environment variable that must be set to true when RequireApproval is enabled or AllowAgentRelease is false.

When either AllowAgentRelease: false or RequireApproval: true is set, the agent must set CHANGESHARP_ALLOW_UNSAFE_RELEASE=true in its environment to proceed with a release. Dry-run previews are always allowed.

Configuration

To use the ChangeSharp MCP server, you need to add it to your AI agent's configuration.

Claude Desktop

Add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "changesharp": {
      "command": "dotnet",
      "args": ["run", "--project", "/absolute/path/to/ChangeSharp.Mcp/ChangeSharp.Mcp.csproj"]
    }
  }
}

Cursor

1. Open Cursor Settings.
2. Go to Features > MCP.
3. Click Add New MCP Server.
4. Name: ChangeSharp
5. Type: command
6. Command: dotnet run --project /absolute/path/to/ChangeSharp.Mcp/ChangeSharp.Mcp.csproj

Benefits

By integrating ChangeSharp with your AI agent: - The agent can automatically create fragments as it implements new features or fixes bugs. - You can ask the agent: "What's the status of our next release?" - The agent can help you prepare a release by summarizing the changes. - Validation is automated before any release is performed.

🧠 Repository Intelligence (Upcoming)

Future versions of the MCP server will include tools to help agents better understand the state of the repository:

• query_changes: Summarizes technical changes (commits, file diffs) since the last release.
• suggest_fragments: Analyzes code changes and suggests the appropriate category and message for a new fragment.
• audit_compliance: Runs the Semantic Safety Gates to ensure the agent's proposed release notes match the actual code modifications.