Skip to content

Latest commit

 

History

History
193 lines (150 loc) · 5.4 KB

mcp.md

File metadata and controls

193 lines (150 loc) · 5.4 KB

Use RepoRipple with coding agents

RepoRipple's MCP server gives coding agents two deterministic, read-only tools:

  • forecast_change: accepts repository-relative paths you plan to edit and predicts their reverse-dependency impact.
  • analyze_worktree: reads the current Git changes, or compares HEAD with a base revision, and returns the actual impact report.

Both tools return typed data containing the changed and impacted files, dependency paths, suggested tests, documentation reminders, risk level, risk reasons, and graph size. No model performs the analysis.

Run from a source checkout

Install uv, clone RepoRipple, and start the stdio server:

git clone https://github.com/adiarora06/RepoRipple.git
cd RepoRipple
uv run reporipple mcp --root /absolute/path/to/your/workspace

An MCP server waits for a client on standard input, so an idle command with no terminal output is expected. Stop it with Ctrl-C when testing manually.

Run the published package with uvx

RepoRipple releases include MCP support on PyPI:

uvx reporipple mcp --root /absolute/path/to/your/workspace

uvx creates and caches an isolated environment, so users do not need to clone the repository or manage a virtual environment.

Codex

Run this from the project you want RepoRipple to inspect. The shell expands $PWD to an absolute allowed root before Codex saves the configuration:

codex mcp add reporipple -- \
  uvx reporipple mcp --root "$PWD"
codex mcp list

Codex CLI and the Codex IDE extension share this configuration. The equivalent ~/.codex/config.toml entry is:

[mcp_servers.reporipple]
command = "uvx"
args = [
  "reporipple", "mcp",
  "--root", "/absolute/path/to/your/workspace"
]

OpenCode v2

Add RepoRipple to the current project:

opencode mcp add reporipple -- \
  uvx reporipple mcp --root .
opencode mcp list

Or add it manually to opencode.jsonc:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "servers": {
      "reporipple": {
        "type": "local",
        "command": [
          "uvx", "reporipple", "mcp", "--root", "."
        ],
        "cwd": "."
      }
    }
  }
}

Cursor

Create .cursor/mcp.json for one project, or ~/.cursor/mcp.json for every project. Replace the root with an absolute path:

{
  "mcpServers": {
    "reporipple": {
      "command": "uvx",
      "args": [
        "reporipple", "mcp",
        "--root", "/absolute/path/to/your/workspace"
      ]
    }
  }
}

Restart Cursor, then confirm that forecast_change and analyze_worktree appear under the RepoRipple server's available tools.

Claude Code

Add RepoRipple at project scope so Claude Code writes a shareable .mcp.json entry:

claude mcp add --scope project --transport stdio reporipple -- \
  uvx reporipple mcp --root .
claude mcp list

Use --scope user instead to make the server available in all of your local projects.

VS Code and GitHub Copilot

Create .vscode/mcp.json in the workspace:

{
  "servers": {
    "reporipple": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "reporipple", "mcp",
        "--root", "${workspaceFolder}"
      ]
    }
  }
}

Run MCP: List Servers, start RepoRipple, and enable its tools in Copilot Chat's Agent mode. Workspace MCP configuration executes local code, so review the command before trusting a repository.

Example prompts

Use RepoRipple to forecast the impact of changing src/billing/store.py and
src/billing/models.py. Explain the dependency paths and tests I should run.
Use RepoRipple to analyze my current worktree. Focus the review on high-risk paths
and missing tests.

For a server whose allowed root contains several repositories, pass the tool's repository argument as a path relative to that root.

Access boundary and privacy

  • --root defaults to the server process's current directory.
  • Requested repositories must resolve to that directory or one of its descendants.
  • Proposed changed files must stay inside the selected repository.
  • Symlinked source files are not scanned.
  • Tools read repository state but never edit files, execute project code, or access the network.
  • Protocol messages use stdout; logs and startup errors use stderr so they cannot corrupt the MCP stream.

The root boundary is enforced by RepoRipple even when an MCP client ignores tool annotations. Give each agent the narrowest useful root.

Client configuration references: Codex MCP, OpenCode MCP, Cursor MCP, Claude Code MCP, and VS Code MCP.

Troubleshooting

  • MCP runtime is unavailable: reinstall RepoRipple so its runtime dependencies are restored.
  • Server starts but prints nothing: this is normal for stdio; inspect it through an MCP client.
  • Repository is outside the allowed root: restart the server with a suitable absolute --root, or request a repository beneath the current root.
  • uvx cannot find RepoRipple: confirm that PyPI is reachable, then retry with uvx --refresh reporipple mcp --root /absolute/path/to/your/workspace or use the source-checkout command above.