CODING AGENT COMPATIBILITY

Choose your coding agent's LinkBunny setup

LinkBunny has one backlink workflow. Your agent reads the shared Skill or uses the local marketplace MCP, inspects your site's content and asks before each action. The setup differs by client: skill directories, MCP formats, private credentials and approval controls.

For a first task without LinkBunny signup, check a launch-directory backlink with your agent. That task uses the free hosted Placement Checker, separately from the paid marketplace.

Start with your local agent

Claude Code

Install /linkbunny, review project-scoped MCP and inspect placement edits in Claude Code.

Codex

Load $linkbunny, merge TOML MCP configuration and follow a request through reporting.

OpenCode

Use command arrays in opencode.json, load the Skill tool and set MCP permission patterns to ask.

GitHub Copilot

Choose local CLI or VS Code, resolve two MCP formats and keep cloud automation out of marketplace actions.

Cursor

Reuse a discovered Skill, configure project MCP and keep Run Modes from automatically approving actions.

Match the client capability to the job

Inspect one existing backlink for free: Use the web checker, or connect a client with remote Streamable HTTP MCP support to https://mcp.getlinkbunny.com/mcp. Its one tool is inspect_placement, with one public source URL and one expected destination per call. No LinkBunny token is needed. It cannot edit a site or handle an exchange.
Prepare a site profile or review a proposed edit: The agent needs access to the site's files and a way to read the full LinkBunny Skill. Use its documented Skill directory below. Read access lets it propose a profile or edit; publication still follows your site's normal process and approval.
Use the paid marketplace through the Skill: The agent also needs approved HTTPS requests, private token access and the ability to pause for each account, profile and placement action. Subscribe first: $20/month covers up to 10 active sites. New profiles need manual marketplace review before matching.
Expose marketplace operations as local MCP tools: The client needs local stdio MCP support; its host needs Node.js 22+ and npm/npx. @linkbunny/mcp@0.1.1 exposes profile creation/update, check-in, reporting and rejection. Pass a private --token-file path, read the guidance and retain per-action confirmation. Server trust alone is not permission for future calls.

Keep unattended clients and cloud agents out of marketplace actions when they cannot pause for your approval. In particular, Copilot's cloud agent runs configured MCP tools without individual approval. Use the interactive Copilot workflow instead. Your coding agent's own account, provider and model costs are separate from LinkBunny.

Find another client's configuration

These official references describe where each client reads skills and MCP settings. Append linkbunny/SKILL.md to the chosen skill directory and install the same canonical file. Review the exact local configuration and native approval controls before enabling marketplace calls. Use the free web checker if those controls are unavailable.

Scroll the table to read configuration and approval details.

Client configuration reference · official sources checked 15 Oct 2026
Client and sources	Skill directory	MCP and credentials	Trust and action controls
Gemini CLI Skills MCP	`.gemini/skills/ or .agents/skills/; personal ~/.gemini/skills/`	`.gemini/settings.json or ~/.gemini/settings.json → mcpServers` Explicit env/headers or supported OAuth. Pass required variables deliberately; sensitive host variables can be filtered.	Keep trust: false and review folder trust. trust: true bypasses tool confirmation.
Antigravity CLI Migration reference	`.agents/skills/; personal ~/.gemini/antigravity-cli/skills/`	`.agents/mcp_config.json or ~/.gemini/config/mcp_config.json → mcpServers` Migration uses env; remote entries use serverUrl. Gemini CLI configuration is not a direct substitute.	Start with a Skill-based profile proposal or the free web checker. Enable marketplace calls only with per-action confirmation in your version.
Cline Skills MCP	`.cline/skills/; personal ~/.cline/skills/`	`CLI: ~/.cline/data/settings/cline_mcp_settings.json → mcpServers; editor: MCP settings UI` Explicit env, remote headers or supported OAuth; keep values outside tracked config.	Leave marketplace tools out of autoApprove. Use Cline's own skill directory rather than assuming .agents support.
Kilo Code Skills MCP	`.kilo/skills/ or .agents/skills/; personal ~/.kilo/skills/`	`kilo.jsonc or .kilo/kilo.jsonc; personal ~/.config/kilo/kilo.jsonc → mcp` Local environment or remote headers/OAuth. Use the current mcp schema rather than older mcpServers examples.	Keep server_tool permission rules asking for confirmation; review agent-level overrides.
Zed Skills MCP and permissions Trust	`.agents/skills/; personal ~/.agents/skills/`	`Zed settings → context_servers; open with zed: open settings file` Local env or remote headers/OAuth. For local LinkBunny use command/args with --token-file.	Trust the worktree after review. Use agent.tool_permissions.default: confirm; inspect per-tool overrides.
Junie CLI Skills MCP	`.junie/skills/ or .agents/skills/; personal ~/.junie/skills/`	`.junie/mcp/mcp.json or ~/.junie/mcp/mcp.json → mcpServers` Explicit env, remote headers or OAuth managed through /mcp.	Use the Skill to prepare a profile first. Keep marketplace MCP disabled unless your active session can pause before each action.
Amp Skills MCP	`.agents/skills/; personal ~/.config/agents/skills/ or ~/.agents/skills/`	`.amp/settings.json or ~/.config/amp/settings.json → amp.mcpServers` Local env/headers or Amp's supported secret storage. Keep LinkBunny's token in a private local file.	Review workspace server approval with amp mcp approve. Server trust is separate from approval for every marketplace action.
Goose Skills Extensions	`.agents/skills/ or .goose/skills/; personal ~/.agents/skills/`	`~/.config/goose/config.yaml → extensions; local cmd/args, remote uri` Extension envs or secret storage; supported remote extensions can use OAuth.	Change the autonomous default to a permission mode that asks before actions. Enabling an extension is not approval for its future calls.
Factory Droid Skills MCP	`.factory/skills/ or .agents/skills/; personal ~/.factory/skills/`	`.factory/mcp.json or ~/.factory/mcp.json → mcpServers` Local env, remote headers or supported OAuth. Keep project config free of secret values.	Review droid mcp permissions, autonomy settings and enterprise overrides; keep marketplace tools asking each time.
Warp Skills MCP	`.agents/skills/; personal ~/.agents/skills/; agent-specific aliases also supported`	`.warp/.mcp.json or ~/.warp/.mcp.json → mcpServers` Explicit env, remote headers or OAuth stored on the device.	Review project-server approval/manual start and global auto-start settings. Keep per-action confirmation for marketplace calls.
Augment / Auggie Skills MCP Permissions	`.augment/skills/ or .agents/skills/; personal ~/.augment/skills/`	`~/.augment/settings.json → mcpServers` Explicit env or remote headers; native integrations have their own browser authorization.	Review toolPermissions in user/project settings. MCP permission names use tool_server; retain confirmation for marketplace actions.
Kiro IDE Skills MCP	`.kiro/skills/; personal ~/.kiro/skills/`	`.kiro/settings/mcp.json or ~/.kiro/settings/mcp.json → mcpServers` Approved env expansion, headers or supported browser OAuth. Review Mcp Approved Env Vars.	Leave marketplace tools out of autoApprove and inspect disabledTools. Use Kiro's skill directory; these paths describe the IDE.
Devin CLI / Desktop Skills MCP	`.agents/skills/ or .devin/skills/; personal ~/.agents/skills/`	`.devin/mcp_config.json or ~/.config/devin/mcp_config.json → mcpServers` Use .devin/mcp_config.local.json for local secret overrides; env/headers or supported OAuth.	Use permissions.ask for mcp__server__tool actions. These MCP files apply from v3000.3; check the migration for older config.json setups.

Keep the same Skill and approval sequence

Use the public canonical Skill, rather than an agent-specific copy of the product rules. The setup prompt uses the agent's supported convention and asks before action gates. Read it before installing and reuse an existing reviewed Skill copy.

Open your coding agent, start a new chat in your site’s project, paste the prompt, and follow its instructions.

For marketplace use, sign in, subscribe, create a token privately, then propose and approve each site profile. Wait for manual review before matching. Approve a check-in, review the contextual nofollow edit, approve publication, then approve reporting separately. Reject an unsuitable request only with approval. No available request means stop, rather than poll.

The API and agent guide owns the workflow; the MCP guide owns the local package. LinkBunny supplies requests and records self-reports. Your agent needs your repository and publishing access, and a placement does not promise rankings, Domain Rating or referral traffic.