The agent connector
Use from your agent
Search, install, update and try Carte Blanche skills and agents without leaving Claude Code, Cursor or Codex.
- Search and read. Your agent finds published, verified skills and agents and reads each one’s license, version and page.
- Install. Your agent gets the files, the folder your tool reads them from (its skills or agents folder), and a SHA-256 for each file, then writes them once you approve. We never write anything on your machine.
- Update. Your agent checks the skills and agents you installed for newer versions.
- Compare on your task. Your agent answers your task once without the skill or agent and once with it, and shows you both. Your task never leaves your machine: this tool takes no task, and nothing about the comparison is sent to us.
- Nothing else. It can’t change anything on Carte Blanche, and it sees only what anyone can see on this site.
Sign in with GitHub, then make a token on your account page. It’s shown once, so copy it then, and keep it like a password. Revoke it there and it stops working at once.
Each token can make 60 calls a minute and 2,000 a day.
The connector is an MCP server at https://carteblanche.com/mcp. Add it to your agent:
Claude Code
In a terminal:
claude mcp add --transport http --scope user carte-blanche https://carteblanche.com/mcp \
--header "Authorization: Bearer YOUR_TOKEN"Replace YOUR_TOKEN with your token. Then run claude mcp list: carte-blanche should say it is connected. Claude Code’s documentation.
Codex
In ~/.codex/config.toml:
[mcp_servers.carte-blanche]
url = "https://carteblanche.com/mcp"
bearer_token_env_var = "CARTE_BLANCHE_TOKEN"Then set CARTE_BLANCHE_TOKEN to your token in your shell profile, and start a new Codex session. Codex’s documentation.
Cursor
In ~/.cursor/mcp.json:
{
"mcpServers": {
"carte-blanche": {
"url": "https://carteblanche.com/mcp",
"headers": { "Authorization": "Bearer ${env:CARTE_BLANCHE_TOKEN}" }
}
}
}Then set CARTE_BLANCHE_TOKEN to your token in your shell profile or system environment, and restart Cursor. Cursor’s documentation.
Checked against each tool’s documentation on 3 November 2026.
- “Find a Carte Blanche skill for code review.”
- “Install the skill name from Carte Blanche.”
- “Check my Carte Blanche skills for updates.”
- “Compare the skill name on my task.”
Your agent asks before it writes a file, as it does for any other change.
~/.claude/skills/Every project. For one project only, use .claude/skills/ inside it.~/.agents/skills/Every project. For one repository only, use .agents/skills/ inside it.~/.cursor/skills/Cursor also reads the Claude Code and Codex folders above.Each skill goes in a folder of its own, named after the skill. Checked against each tool’s documentation on 3 November 2026.
~/.claude/agents/Every project. For one project only, use .claude/agents/ inside it. It gives the agent the tools it asks for, and no others.~/.cursor/agents/Cursor also reads the Claude Code folder above. It has no tools field, so it gives every agent every tool, whatever the agent asks for.Each agent is one Markdown file, with its license file beside it under a name that doesn’t end in .md. Checked against each tool’s documentation on 6 November 2026.
A small command-line tool will install, update, list and remove skills the same way, keep a record of what it installed, and refuse to write any file that doesn’t match its hash. We’ll put the install command here once it’s published.
For each call: which token made it, which tool, which skill, when, and whether it worked. Never the words you search for or anything you’re working on. We keep these records 30 days. We also count how many times each skill is installed or updated each day, with nothing about who did it. See Privacy.