Use LinkBunny with local GitHub Copilot

Keep marketplace actions in an interactive local session

GitHub's cloud agent uses configured MCP tools autonomously, without asking for approval before each call. Do not give it the local LinkBunny marketplace server or token. LinkBunny requires approval for registration, check-in, publication, reporting and rejection; cloud tool execution cannot supply those pauses. Use a local CLI or IDE session where you can review the proposed action.

Copilot cloud tool behavior · checked 15 Oct 2026: GitHub cloud MCP configuration

Install the shared LinkBunny Skill

Review the canonical Skill, then use the setup prompt or the manual command below. The Skill holds the API workflow and approval rules. Local MCP is optional; both interfaces follow the same instructions.

In bash or zsh, this downloads the Skill to ~/.agents/skills/linkbunny/SKILL.md and refuses to overwrite a copy. It needs curl with --remove-on-error support. For one repository, save the reviewed file at .agents/skills/linkbunny/SKILL.md instead. Keep credentials out of the Skill.

skill_dir="$HOME/.agents/skills/linkbunny"
skill_file="$skill_dir/SKILL.md"
mkdir -p "$skill_dir" &&
if [ -e "$skill_file" ]; then
  printf 'Refusing to overwrite %s\n' "$skill_file" >&2; exit 1
else
  curl -fsSL --remove-on-error "https://getlinkbunny.com/skill.md" -o "$skill_file"
fi

Copilot CLI reads project skills in .github/skills or .agents/skills, and personal skills in ~/.copilot/skills or ~/.agents/skills.

Use /skills reload, then /skills info linkbunny in an interactive session. Ask it to use /linkbunny and read the whole Skill. Reuse a reviewed copy rather than adding the same skill in each directory.

Subscribe before creating a marketplace token

$20/month covers up to 10 active sites. Sign in to Account, subscribe, then create a token. Approve account, subscription and credential actions as they arise. Your agent's account and model usage are separate from LinkBunny's plan.

Store LINKBUNNY_API_TOKEN in an approved secret store or a private, Git-ignored file. The local MCP examples use --token-file: only the absolute file path goes in configuration. The loader reads that variable without executing the file as a shell script. Keep the value out of prompts, command arguments, source, URLs, logs and screenshots.

Have the agent propose a profile from your site's content before registration. Each new profile needs manual marketplace review before matching. Approve one check-in at a time; check_in can reserve an assignment. Review each contextual nofollow edit, approve publication, then approve reporting separately. Approve rejection when no suitable context exists. Stop when review is pending or no request is available; do not poll automatically.

The agent workflow owns these steps. The directory backlink tutorial uses the free hosted checker for an existing link and needs no LinkBunny signup or token.

Configure Copilot CLI with .mcp.json

With Node.js 22+ and npm/npx on the CLI host, review and approve this merge into the site's .mcp.json. Replace the absolute path with your private token file. Inspect an existing Claude Code entry first; Copilot CLI may already read that same file.

Prepare project-scoped registration for the local stdio server @linkbunny/mcp@0.1.1 in this coding agent. Follow https://getlinkbunny.com/docs/agents#mcp for the correct client configuration, secret loading, trust, restart, and safe verification. Never reveal or copy the LinkBunny API token into chat, command arguments, logs, screenshots, or tracked configuration. Before changing persistent MCP configuration, show me the exact client, scope, config diff, pinned command, secret-loading path, and restart/trust steps, then ask for approval. After setup, read linkbunny://guidance/skill before using any LinkBunny tool. Do not call a LinkBunny tool merely to test setup.

{
  "mcpServers": {
    "linkbunny": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "@linkbunny/mcp@0.1.1",
        "--token-file",
        "/absolute/project/.env.linkbunny"
      ],
      "tools": [
        "*"
      ]
    }
  }
}

Copilot CLI also reads .github/mcp.json and the personal ~/.copilot/mcp-config.json. Project definitions take precedence; a closer directory can override the same server name. Project servers load after folder trust. Copilot CLI does not read .vscode/mcp.json directly.

Only PATH is inherited automatically by local MCP servers. The --token-file argument avoids relying on an inherited token or putting its value in JSON. tools: ["*"] makes the server's tools available; it is a capability filter, not approval to run them. Keep --allow-all, --yolo and persistent approvals for marketplace tools off.

copilot mcp list --json
copilot mcp get linkbunny --json

Inspect the configured command, scope and path privately. A saved entry alone does not prove a server connection. In an interactive session, use /mcp to inspect the active server and available tools. Read the installed Skill first; don't invoke check_in or another marketplace tool to test setup.

Copilot CLI configuration, trust and skills · checked 15 Oct 2026: MCP setup · Skills · CLI permissions

Match VS Code's MCP format to the session

VS Code's editor format uses .vscode/mcp.json with a top-level servers object. The portable Copilot format above uses .mcp.json with mcpServers. Copilot harness sessions read the portable format directly. VS Code can also forward editor-format servers to the session, except entries requiring interactive input such as ${input:...} variables.

Use one definition for this server and inspect the active session's tool list. Follow the existing VS Code setup for the editor format. Review workspace and server trust prompts, and choose individual tool approvals. If a workspace or organization policy prevents those controls, use the free web checker for placement checks and leave marketplace tools disabled.

VS Code formats and session forwarding · checked 15 Oct 2026: MCP configuration reference · Manage MCP servers

When Copilot shows the wrong server or no tools

The CLI skips a project server

Start an interactive session in the intended repository and review folder trust. Check .mcp.json before .github/mcp.json and user configuration. Don't bypass trust through a headless prompt or broaden permissions to make it load.

The token exists in your shell but MCP cannot read it

Check the private file on the CLI host and its absolute path. Copilot's MCP environment is not your entire shell environment. After an approved token rotation, restart the server. Keep diagnostic output private.

The CLI entry works differently from the editor entry

Check mcpServers versus servers, configuration precedence and the selected harness. Editor input variables cannot be forwarded to a Copilot harness. Your Copilot account/provider and LinkBunny subscription are separate requirements.

Choose another coding agent's setup.