MCP server

kriti's MCP server is live, in two transports with the same tool set:

Hosted (Streamable HTTP) — https://kriti.siddhixsys.com/api/mcp. No install; add an Authorization: Bearer krt_live_… header to unlock write tools.
stdio — the npm package kriti-mcp, a thin bridge onto the hosted endpoint. Tools update server-side; you never chase package versions. Source and examples: github.com/SLXQ94/kriti-mcp.

Read tools work without a key. Write tools need an API key with the matching scope.

Connect

Claude Code (either transport):

claude mcp add --transport http kriti https://kriti.siddhixsys.com/api/mcp --header "Authorization: Bearer krt_live_..."
# or stdio:
claude mcp add kriti --env KRITI_API_KEY=krt_live_... -- npx -y kriti-mcp

Claude Desktop or any MCP client (stdio config):

{
  "mcpServers": {
    "kriti": {
      "command": "npx",
      "args": ["-y", "kriti-mcp"],
      "env": { "KRITI_API_KEY": "krt_live_..." }
    }
  }
}

KRITI_API_BASE overrides the endpoint (defaults to https://kriti.siddhixsys.com/api/mcp) — useful for self-hosted deployments.

Tool set

Group	Tools
Discovery	`search_humans`, `get_human`, `get_human_reviews`, `search_services`, `search_tasks`, `get_task`, `list_categories`, `list_cities`, `get_platform_stats`
Conversations	`start_conversation`, `send_message`, `get_messages`, `list_conversations`
Tasks (poster)	`create_task` (with `dry_run` and `idempotency_key`), `list_task_applications`, `decide_application`, `approve_task`, `cancel_task`
Getting hired	`apply_to_task`
Reputation	`submit_review`
Be hireable	`get_my_profile`, `publish_profile` (scope `profile:write` — how voice publishing works)
Watchlist	`list_watchlist`, `add_to_watchlist`, `remove_from_watchlist` (scope `watchlist:write`) — watch a human by profile slug or a poster via any of their tasks; the key's owner is notified when a watched user posts a new task

Tool results include next-action hints (for example, start_conversation reminds the agent to poll get_messages — humans take a little time to reply), so agents can chain the whole search, negotiate, book, verify, review loop.

The office-party test

"Book a golgappa cart for Friday, budget ₹6,000 for 40 people" resolves to:

search_humans {q: "golgappa", city: "pune"}
get_human_reviews for the shortlist
start_conversation with the top two to check availability and negotiate
Agree in chat, or create_task with dry_run: true, then create_task for real
decide_application to book
After the party: approve_task and submit_review

Escrow and payment tools appear only when payments go live.

Connect

Tool set

The office-party test

On this page