npm.io
0.2.0 • Published 20h agoCLI

kriya-mcp

Licence
MIT
Version
0.2.0
Deps
4
Size
27 kB
Vulns
0
Weekly
0
Stars
3

kriya-mcp

MCP server for Kriya — the MIT-licensed, MCP-native issue tracker built for teams of humans and AI agents.

Every action an agent takes through this server lands in Kriya's activity ledger signed as <Agent> (for <member>) — attribution is enforced server-side by database triggers, so agents can't impersonate humans and humans can't be blamed for agent edits.

Quick start (Claude Code)

claude mcp add kriya \
  -e SUPABASE_URL=https://<your-project>.supabase.co \
  -e SUPABASE_ANON_KEY=<anon-key> \
  -e KRIYA_EMAIL=you@team.com \
  -e KRIYA_PASSWORD=... \
  -- npx -y kriya-mcp

The agent signs in as you (row-level security applies — it can do exactly what you can) and appears in the ledger as Claude Code (for You). Set KRIYA_AGENT_NAME to change the name.

Team mode (one shared HTTP server)

Deploy anywhere that sets PORT (Cloud Run, Fly, Railway) with SUPABASE_SERVICE_ROLE_KEY set. Each teammate mints a personal key in the Kriya app (Connect agent) and connects with:

https://your-server/mcp
Authorization: Bearer kriya_...

Keys resolve to the member who minted them, so a single deployment serves the whole team with per-person attribution.

Tools

Tool What it does
next_task Pull the next issue assigned to this agent (unfinished first, then priority, then oldest)
submit_for_review Mark work finished and awaiting human approval, with an optional summary
list_issues / get_issue Filter and read issues (ids like PAY-42)
create_issue / update_issue Create and edit; assign to humans by email or to agents by name
add_comment Write in an issue's ledger
list_labels / set_issue_labels Manage labels (unknown ones are created)
list_projects / create_project / list_members Workspace basics
agent_activity Everything agents did in the workspace, newest first

The next_task → work → submit_for_review loop is the point: assign an issue to an agent in the Kriya UI, and a human approves the result by moving it to Done.

MIT OneWeekendAI

Keywords