Works inside your coding agent: how FoxPlug reads Claude Code, Cline, Cursor, Gemini CLI, Grok Build, Kiro, OpenAI Codex

Claude Code

Claude Code is an AI coding tool that runs in your terminal. FoxPlug adds a small hook to it that sends a short summary each time a session stops.

What it reads

• A short summary of what the session did: plain text only, up to about 8 KB, sent by the hook when the session stops. Anything longer is refused, and a raw transcript is never accepted or stored
• The area, the version and up to 20 file names: optional extras the hook can add: the part of the product worked on, a version label and the names of changed files (200 characters each). No file contents
• What is stored from it: one update per summary, titled with the summary's first line, with the rest as its description (up to 4,000 characters), after anything that looks like a secret has been replaced. The same summary sent twice in one day is stored once

Moments

• A Claude Code session stops and sends its summary. You finish a session that added a pricing toggle. The hook sends a paragraph about it. FoxPlug scores it like any other update and the brain may offer it as a say card. Nothing is posted.

Set up

1. Open FoxPlug, go to Settings, then Your code, then Read straight from the AI tool you code in, and make a Claude Code key. Copy the one line it gives you; the key is shown once.
2. Paste that line in the terminal where Claude Code runs. It adds a Stop hook to Claude Code's settings and keeps every setting you already had.
3. Finish a session in Claude Code. The summary arrives in FoxPlug as an update; nothing is posted until you approve it.

What it never reads

• Your code, your files or your conversation: the hook sends only the short summary
• Anything on a timer: nothing arrives unless a session stops

How it differs

The other coding tools here connect through FoxPlug's MCP address with an agent key and call it themselves. Claude Code uses a hook that sends a summary when a session stops, with its own ingest key.

Cline

Cline is an AI coding tool that can call outside tool servers over MCP, so it can tell FoxPlug what it just built.

What it reads

• A short summary the tool sends after a change: one paragraph of plain text, an optional note on decisions, whether a user would notice the change, and the project name; sent when the tool calls the log update tool
• A request for your brief or your drafts: the tool can ask what to post this week and list the drafts that are ready; those calls record nothing

Moments

• The tool calls log update after a change a user would notice. You finish a change and the tool sends a one-paragraph summary. FoxPlug scores it like any other update and the brain may offer it as a say card. Nothing is posted.

Set up

1. Open FoxPlug, go to Settings, then Your code, then Read straight from the AI tool you code in, and click Make an agent key. Copy it now; it is shown once.
2. Put the block below in cline mcp settings.json.
3. Add the one line below to Cline rules so Cline tells FoxPlug what it built.

Config file: cline_mcp_settings.json (in the Cline panel: MCP Servers, Configure, Configure MCP Servers; the CLI keeps it at ~/.cline/data/settings/cline_mcp_settings.json) (JSON). Replace fpk_YOUR_AGENT_KEY with your own agent key.

{
  "mcpServers": {
    "foxplug": {
      "type": "streamableHttp",
      "url": "https://foxplug.com/mcp",
      "headers": {
        "Authorization": "Bearer fpk_YOUR_AGENT_KEY"
      }
    }
  }
}

Or from a terminal: cline mcp add foxplug https://foxplug.com/mcp --transport http --header "Authorization: Bearer fpk_YOUR_AGENT_KEY" --yes

Then add this line to the rules file the tool reads (for example AGENTS.md), so it tells FoxPlug what it built:

After you finish a change a user would notice, call the foxplug log_update tool with a one-paragraph plain summary. For plumbing nobody outside needs to hear about (refactors, dependency bumps, migrations), call it with user_facing set to false.

What it never reads

• Your code, your files or your chat with the tool: it sends only the paragraph it is asked to send
• Anything on a timer: nothing arrives unless the tool calls FoxPlug

How it differs

Unlike the tools FoxPlug reads when you press Check for what's new, Cline sends FoxPlug a summary the moment it finishes, because you told it to; FoxPlug never fetches anything from it.

Cursor

Cursor is an AI coding tool that can call outside tool servers over MCP, so it can tell FoxPlug what it just built.

What it reads

• A short summary the tool sends after a change: one paragraph of plain text, an optional note on decisions, whether a user would notice the change, and the project name; sent when the tool calls the log update tool
• A request for your brief or your drafts: the tool can ask what to post this week and list the drafts that are ready; those calls record nothing

Moments

• The tool calls log update after a change a user would notice. You finish a change and the tool sends a one-paragraph summary. FoxPlug scores it like any other update and the brain may offer it as a say card. Nothing is posted.

Set up

1. Open FoxPlug, go to Settings, then Your code, then Read straight from the AI tool you code in, and click Make an agent key. Copy it now; it is shown once.
2. Put the block below in.cursor/mcp.json in the project, or ~/.cursor/mcp.json for every project.
3. Add the one line below to Cursor rules so Cursor tells FoxPlug what it built.

Config file: .cursor/mcp.json in the project, or ~/.cursor/mcp.json for every project (JSON). Replace fpk_YOUR_AGENT_KEY with your own agent key.

{
  "mcpServers": {
    "foxplug": {
      "url": "https://foxplug.com/mcp",
      "headers": {
        "Authorization": "Bearer fpk_YOUR_AGENT_KEY"
      }
    }
  }
}

Then add this line to the rules file the tool reads (for example AGENTS.md), so it tells FoxPlug what it built:

After you finish a change a user would notice, call the foxplug log_update tool with a one-paragraph plain summary. For plumbing nobody outside needs to hear about (refactors, dependency bumps, migrations), call it with user_facing set to false.

What it never reads

• Your code, your files or your chat with the tool: it sends only the paragraph it is asked to send
• Anything on a timer: nothing arrives unless the tool calls FoxPlug

How it differs

Unlike the tools FoxPlug reads when you press Check for what's new, Cursor sends FoxPlug a summary the moment it finishes, because you told it to; FoxPlug never fetches anything from it.

Gemini CLI

Gemini CLI is an AI coding tool that can call outside tool servers over MCP, so it can tell FoxPlug what it just built.

What it reads

• A short summary the tool sends after a change: one paragraph of plain text, an optional note on decisions, whether a user would notice the change, and the project name; sent when the tool calls the log update tool
• A request for your brief or your drafts: the tool can ask what to post this week and list the drafts that are ready; those calls record nothing

Moments

• The tool calls log update after a change a user would notice. You finish a change and the tool sends a one-paragraph summary. FoxPlug scores it like any other update and the brain may offer it as a say card. Nothing is posted.

Set up

1. Open FoxPlug, go to Settings, then Your code, then Read straight from the AI tool you code in, and click Make an agent key. Copy it now; it is shown once.
2. Put the block below in ~/.gemini/settings.json, or.gemini/settings.json in the project.
3. Add the one line below to the file where you keep standing instructions for Gemini CLI so Gemini CLI tells FoxPlug what it built.

Config file: ~/.gemini/settings.json, or .gemini/settings.json in the project (JSON). Replace fpk_YOUR_AGENT_KEY with your own agent key.

{
  "mcpServers": {
    "foxplug": {
      "httpUrl": "https://foxplug.com/mcp",
      "headers": {
        "Authorization": "Bearer fpk_YOUR_AGENT_KEY"
      }
    }
  }
}

Or from a terminal: gemini mcp add --transport http --header "Authorization: Bearer fpk_YOUR_AGENT_KEY" foxplug https://foxplug.com/mcp (writes the project file; add -s user for ~/.gemini/settings.json)

Then add this line to the rules file the tool reads (for example AGENTS.md), so it tells FoxPlug what it built:

After you finish a change a user would notice, call the foxplug log_update tool with a one-paragraph plain summary. For plumbing nobody outside needs to hear about (refactors, dependency bumps, migrations), call it with user_facing set to false.

What it never reads

• Your code, your files or your chat with the tool: it sends only the paragraph it is asked to send
• Anything on a timer: nothing arrives unless the tool calls FoxPlug

How it differs

Unlike the tools FoxPlug reads when you press Check for what's new, Gemini CLI sends FoxPlug a summary the moment it finishes, because you told it to; FoxPlug never fetches anything from it.

Grok Build

Grok Build is an AI coding tool that can call outside tool servers over MCP, so it can tell FoxPlug what it just built.

What it reads

• A short summary the tool sends after a change: one paragraph of plain text, an optional note on decisions, whether a user would notice the change, and the project name; sent when the tool calls the log update tool
• A request for your brief or your drafts: the tool can ask what to post this week and list the drafts that are ready; those calls record nothing

Moments

• The tool calls log update after a change a user would notice. You finish a change and the tool sends a one-paragraph summary. FoxPlug scores it like any other update and the brain may offer it as a say card. Nothing is posted.

Set up

1. Open FoxPlug, go to Settings, then Your code, then Read straight from the AI tool you code in, and click Make an agent key. Copy it now; it is shown once.
2. Run the command below in your terminal.
3. Add the one line below to AGENTS.md so Grok Build tells FoxPlug what it built.

Config file: ~/.grok/config.toml, or the grok mcp add command (add --scope project to write .grok/config.toml in the project) (command line, or TOML). Replace fpk_YOUR_AGENT_KEY with your own agent key.

grok mcp add --transport http foxplug https://foxplug.com/mcp --header "Authorization: Bearer fpk_YOUR_AGENT_KEY"

Or from a terminal: [mcp_servers.foxplug] url = "https://foxplug.com/mcp" headers = { "Authorization" = "Bearer fpk_YOUR_AGENT_KEY" }

Then add this line to the rules file the tool reads (for example AGENTS.md), so it tells FoxPlug what it built:

After you finish a change a user would notice, call the foxplug log_update tool with a one-paragraph plain summary. For plumbing nobody outside needs to hear about (refactors, dependency bumps, migrations), call it with user_facing set to false.

What it never reads

• Your code, your files or your chat with the tool: it sends only the paragraph it is asked to send
• Anything on a timer: nothing arrives unless the tool calls FoxPlug

How it differs

Unlike the tools FoxPlug reads when you press Check for what's new, Grok Build sends FoxPlug a summary the moment it finishes, because you told it to; FoxPlug never fetches anything from it.

Kiro

Kiro is an AI coding tool that can call outside tool servers over MCP, so it can tell FoxPlug what it just built.

What it reads

• A short summary the tool sends after a change: one paragraph of plain text, an optional note on decisions, whether a user would notice the change, and the project name; sent when the tool calls the log update tool
• A request for your brief or your drafts: the tool can ask what to post this week and list the drafts that are ready; those calls record nothing

Moments

• The tool calls log update after a change a user would notice. You finish a change and the tool sends a one-paragraph summary. FoxPlug scores it like any other update and the brain may offer it as a say card. Nothing is posted.

Set up

1. Open FoxPlug, go to Settings, then Your code, then Read straight from the AI tool you code in, and click Make an agent key. Copy it now; it is shown once.
2. Put the block below in.kiro/settings/mcp.json in the workspace, or ~/.kiro/settings/mcp.json for every workspace (merged, workspace wins).
3. Add the one line below to your Kiro steering file so Kiro tells FoxPlug what it built.

Config file: .kiro/settings/mcp.json in the workspace, or ~/.kiro/settings/mcp.json for every workspace (merged, workspace wins) (JSON). Replace fpk_YOUR_AGENT_KEY with your own agent key.

{
  "mcpServers": {
    "foxplug": {
      "url": "https://foxplug.com/mcp",
      "headers": {
        "Authorization": "Bearer fpk_YOUR_AGENT_KEY"
      }
    }
  }
}

Then add this line to the rules file the tool reads (for example AGENTS.md), so it tells FoxPlug what it built:

After you finish a change a user would notice, call the foxplug log_update tool with a one-paragraph plain summary. For plumbing nobody outside needs to hear about (refactors, dependency bumps, migrations), call it with user_facing set to false.

What it never reads

• Your code, your files or your chat with the tool: it sends only the paragraph it is asked to send
• Anything on a timer: nothing arrives unless the tool calls FoxPlug

How it differs

Unlike the tools FoxPlug reads when you press Check for what's new, Kiro sends FoxPlug a summary the moment it finishes, because you told it to; FoxPlug never fetches anything from it.

OpenAI Codex

Codex CLI is an AI coding tool that can call outside tool servers over MCP, so it can tell FoxPlug what it just built.

What it reads

• A short summary the tool sends after a change: one paragraph of plain text, an optional note on decisions, whether a user would notice the change, and the project name; sent when the tool calls the log update tool
• A request for your brief or your drafts: the tool can ask what to post this week and list the drafts that are ready; those calls record nothing

Moments

• The tool calls log update after a change a user would notice. You finish a change and the tool sends a one-paragraph summary. FoxPlug scores it like any other update and the brain may offer it as a say card. Nothing is posted.

Set up

1. Open FoxPlug, go to Settings, then Your code, then Read straight from the AI tool you code in, and click Make an agent key. Copy it now; it is shown once.
2. Set the key as an environment variable: export FOXPLUG_AGENT_KEY="fpk_YOUR_AGENT_KEY" (PowerShell: setx FOXPLUG_AGENT_KEY "fpk_YOUR_AGENT_KEY", then open a new terminal)
3. Put the block below in ~/.codex/config.toml (or.codex/config.toml in a trusted project).
4. Add the one line below to AGENTS.md so Codex CLI tells FoxPlug what it built.

Config file: ~/.codex/config.toml (or .codex/config.toml in a trusted project) (TOML). Replace fpk_YOUR_AGENT_KEY with your own agent key.

[mcp_servers.foxplug]
url = "https://foxplug.com/mcp"
bearer_token_env_var = "FOXPLUG_AGENT_KEY"

Then add this line to the rules file the tool reads (for example AGENTS.md), so it tells FoxPlug what it built:

After you finish a change a user would notice, call the foxplug log_update tool with a one-paragraph plain summary. For plumbing nobody outside needs to hear about (refactors, dependency bumps, migrations), call it with user_facing set to false.

What it never reads

• Your code, your files or your chat with the tool: it sends only the paragraph it is asked to send
• Anything on a timer: nothing arrives unless the tool calls FoxPlug

How it differs

Unlike the tools FoxPlug reads when you press Check for what's new, Codex CLI sends FoxPlug a summary the moment it finishes, because you told it to; FoxPlug never fetches anything from it.