The value is the constraint: the model has to name what it is doing at each step — recognise, examine, integrate, transcend — instead of producing one confident paragraph. The journey it returns is readable afterwards, so you can see where an answer turned, and the tool only releases the model to speak once it reports WISDOM_READY.
A reasoning-scaffold server built on the Lotus Sutra framework: the model works a problem step by step, each step tagged with a technique, until the tool reports the contemplation complete.
- A stepwise process where each step carries a tag from one of the wisdom domains — skillful means, non-dual recognition, meta-cognitive, process flow — and the first begin step returns the full framework
- Meditation steps that pause deliberately to let an insight settle, with a duration between 1 and 10 seconds
- Journey tracking: both the tag path and the movement between wisdom domains, retrievable as a summary of every step taken
- Explicit statuses to drive the loop — FRAMEWORK_RECEIVED on the opening step, MEDITATION_COMPLETE on a pause, WISDOM_READY when the process is done
- Structured output alongside the text response, validated against the tool's own output schema
- Two prompts, contemplate and deep-inquiry, that set up a guided session in one step
- An interactive inline trace in clients supporting MCP ext-apps, following the host's light or dark theme and reachable by keyboard
- Behavioural annotations on the tool — readOnlyHint, idempotentHint, destructiveHint false, openWorldHint false — so a host can treat the call as side-effect free
Nothing to authorize. Published on npm as lotus-wisdom-mcp for stdio use, and also hosted as a remote streamable-http endpoint at https://lotus-wisdom-mcp.linxule.workers.dev/mcp. The hosted one keeps no session state, so pass the previousJourney string back to carry a contemplation forward.
One command — npx -y lotus-wisdom-mcp