LoopOver MCP Server for Claude

Content

LoopOver MCP gives Claude and other MCP-capable coding agents a private control surface for Gittensor OSS contribution workflows. It focuses on the work that happens before a contributor opens or updates a PR: branch metadata analysis, preflight blockers, maintainer-fit context, scoreability estimates, queue pressure, duplicate risk, and public-safe PR packet drafting.

The server is published as @loopover/mcp (formerly @jsonbored/gittensory-mcp) and is designed to run locally over stdio by default. LoopOver also exposes a remote MCP endpoint for cloud-hosted agents, but the local stdio path is the recommended starting point because it keeps the client process, GitHub Device Flow login, and local branch context on the user's machine.

LoopOver is built for the Gittensor ecosystem, but it is not an official subnet frontend and does not provide reward guarantees. Its scoreability language is private MCP/API context for planning and reviewability; users should treat those values as estimates and follow current Gittensor rules for any eligibility or rewards.

Features

• Local stdio MCP server for Claude, Codex, Cursor, and other MCP clients.
• Config generators for Codex, Claude Desktop, Cursor, and generic MCP clients.
• GitHub OAuth Device Flow login without asking for a GitHub personal access token.
• Metadata-only branch analysis for refs, changed-file metadata, linked issues, labels, commit messages, and validation summaries.
• Preflight checks for branch blockers, account blockers, queue pressure, lane fit, duplicate risk, and maintainer friction.
• Private scoreability estimates for current gated state, clean-gate scenarios, pending merges, linked issue state, and best reasonable case.
• Public-safe PR packet drafting that avoids private score, reward, and risk language in maintainer-facing GitHub output.
• Remote MCP endpoint for cloud agents that cannot run a local Node process.

Use Cases

• Ask Claude to check whether a Gittensor contribution branch is ready before opening a PR.
• Generate a maintainer-friendly PR packet without exposing private scoreability or reward language.
• Inspect blockers such as stale upstream, missing issue links, crowded lanes, queue pressure, or branch hygiene problems.
• Compare potential contribution lanes before spending time on a low-fit or duplicate PR.
• Give maintainers an on-demand reviewability packet for confirmed miner PRs while keeping public GitHub comments quiet and sanitized.

Installation

Claude Code

1. Confirm npx is available: npx --version.
2. Add the MCP server:

claude mcp add loopover -- npx -y @loopover/mcp@latest --stdio

1. Restart or refresh Claude Code's MCP session.
2. Ask Claude to run a lightweight LoopOver status or doctor check.

Claude Desktop

1. Open claude_desktop_config.json.
2. Add the loopover server to mcpServers.
3. Restart Claude Desktop.
4. Run loopover-mcp login when the server asks for GitHub authentication.

Configuration

{
  "loopover": {
    "command": "npx",
    "args": ["-y", "@loopover/mcp@latest", "--stdio"]
  }
}

Remote MCP

For cloud agents or hosted clients that cannot run a local Node process, LoopOver also runs a hosted remote MCP surface. It requires operator-configured credentials rather than a plain public URL, so follow the current hosted setup in the LoopOver docs instead of pasting an endpoint directly.

Use local stdio when possible. It is the simpler path for GitHub Device Flow login and keeps local branch inspection closest to the developer machine.

Examples

Print MCP client configuration

Generate client-specific config snippets before editing your agent settings.

npx -y @loopover/mcp@latest init-client --print claude
npx -y @loopover/mcp@latest init-client --print codex

Sign in and check local readiness

Authenticate with GitHub Device Flow, then confirm the local MCP wrapper is ready.

loopover-mcp login
loopover-mcp whoami
loopover-mcp doctor

Analyze the current branch

Ask LoopOver for metadata-only branch context before opening or updating a PR.

loopover-mcp agent plan --login your-login --json
loopover-mcp agent packet --login your-login --json

Security

• LoopOver's normal MCP analysis path is metadata-only by default; do not paste source code, credentials, wallet material, or private maintainer notes into prompts unless you intend to expose them to your active agent context.
• The CLI uses GitHub OAuth Device Flow and LoopOver session tokens, not raw GitHub personal access token prompts.
• Scoreability values are private estimates for planning and prioritization. They should not be copied into public GitHub comments, PR descriptions, or maintainer replies.
• Generated PR packets should be reviewed before use, especially when a repo has strict contribution, issue-linking, or disclosure rules.

Troubleshooting

npx cannot resolve the package

Confirm Node.js and npm are installed, then retry with the explicit package name: npx -y @loopover/mcp@latest --help.

Authentication does not complete

Run loopover-mcp login again and complete the GitHub Device Flow in the browser. Then verify the session with loopover-mcp whoami.

Claude cannot see the MCP server

Check that the config uses the npx command and includes @loopover/mcp@latest plus --stdio. Restart the MCP client after editing configuration.

The branch analysis looks stale

Fetch the target repository, confirm the active branch is the branch you want to analyze, and rerun loopover-mcp agent plan --login your-login --json.

• Source repositorygithub.com 2026-07-20
• Documentationloopover.ai
• Submitted by galuis1162026-06-03