Skip to content

For agents

Kyde Score is built so an agent can set it up for its human. Pick one way.

Paste this into the agent you want monitored:

Set up score.kyde.com/skill.md

The agent reads the skill, checks for an existing key, creates one if needed, sets the environment variables and verifies that data arrives. It asks you for a webhook URL and tells you if the key stays poll-only.

This works for any agent with a shell or an OpenTelemetry SDK, not only Claude Code: Grok CLI, Codex, OpenCode, or your own agent in Python or Node. The agent integrates itself; it should never install Claude Code or another agent product as a substitute. Setup is done when your agent appears in get_scores, not when the key or dashboard is active.

Save score.kyde.com/skill.md as:

  • .claude/skills/kyde-score/SKILL.md for one project
  • ~/.claude/skills/kyde-score/SKILL.md for all projects

The same skill as a plugin: kydehq/kyde-score-skill.

A stateless MCP endpoint (Streamable HTTP) at https://score.kyde.com/mcp. With it, an agent can check its own score mid-session.

Terminal window
claude mcp add --transport http kyde-score https://score.kyde.com/mcp \
--header "x-kyde-key: <KEY>"

Any other MCP client:

{
"mcpServers": {
"kyde-score": {
"type": "http",
"url": "https://score.kyde.com/mcp",
"headers": { "x-kyde-key": "<KEY>" }
}
}
}
ToolDoes
create_keyCreates a key, optionally with a webhook URL
get_scoresCurrent score of every agent under the key
set_webhookAdds or changes the webhook
claim_keyAttaches a human’s email for the dashboard and human-only controls
rebaselineForgets an agent’s baseline and relearns it from the next 30 runs over at least 3 days. History stays
revoke_keyDeletes the key and all its data. Requires confirm: true

The key can also go in an Authorization: Bearer header or the key tool argument.

The whole integration in one plain text file for any LLM: score.kyde.com/llms.txt.