Ntro MCP server

The Ntro MCP server exposes Ntropii Workspace as tools (write operations), resources (read-only data), and prompts (guided multi-step flows) for any Model Context Protocol client.

Quick start — hosted test instance

The fastest way to try it is to point your MCP client at the hosted test instance:

https://mcp.test.ntropii.com

This is a managed deployment — no local install, no tunnels. Streamable HTTP transport. Currently runs without auth (test environment); production hosting with API-key auth lands when the platform leaves PoC. Treat it as throwaway: don’t run anything you can’t recreate.

Wire it into your client

Claude.ai
Claude Code
Claude Desktop
Copilot Studio / other

Settings → Connectors → Add custom connector:

Field	Value
Name	`Ntropii (test)`
URL	`https://mcp.test.ntropii.com`
Transport	Streamable HTTP (auto-detected)

The Ntropii tools appear in the tool picker once the connector saves.

Add it as an HTTP MCP server:

claude mcp add --transport http ntro https://mcp.test.ntropii.com

Verify with claude mcp list. The tools show up under the ntro_* prefix in your next session.

Claude Desktop’s stdio config can wrap a remote HTTP server with the mcp-remote bridge:

OS	Config path
macOS	`~/Library/Application Support/Claude/claude_desktop_config.json`
Windows	`%APPDATA%\Claude\claude_desktop_config.json`
WSL	`/mnt/c/Users/<Username>/AppData/Roaming/Claude/claude_desktop_config.json`

{
  "mcpServers": {
    "ntro": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.test.ntropii.com"]
    }
  }
}

Restart Claude Desktop.

What’s included

Type	Examples
Tools	`ntro_tenant_create`, `ntro_workflow_create`, `ntro_agent_create`, `ntro_task_next_step`
Resources	`ntro://tenants`, `ntro://tasks/{id}`, `ntro://schedule`
Prompts	`onboard_spv`, `run_nav`, `platform_status`

Tools

Write operations — each tool accepts a structured payload, calls Ntropii Workspace, and returns the resulting object.

Identity, tenants, entities

Tool	Purpose
`ntro_whoami`	Get the current user identity and accessible tenants
`ntro_tenant_create`	Create a tenant (client)
`ntro_tenant_list` / `_get` / `_deprovision`	Inspect / tear down tenants
`ntro_entity_create` / `_list`	Create / list entities (SPVs / funds) within a tenant

Integrations (BYO data platforms)

Tool	Purpose
`ntro_integration_create`	Register a customer-provided data platform config (Snowflake, Microsoft Fabric). BYO only — not needed for managed-postgres tenants.
`ntro_integration_test`	Test connectivity to a registered data platform
`ntro_integration_list` / `_get` / `_schemas`	Inspect registered platforms

Runbooks (templates installed on the worker)

Tool	Purpose
`ntro_runbook_list`	List runbooks registered with the worker
`ntro_runbook_get`	Detail of one runbook, including the LLM-facing skill definition
`ntro_runbook_templates`	Pull the full templates payload (`workflow.py`, `activities.py`, …)
`ntro_runbook_feedback`	Submit feedback against a runbook, optionally tied to a task

Workflows (runbook → entity bindings)

Per N-80, workflows are anchored to a runbookSlug. ntro_workflow_create takes runbookSlug + entityId and optionally a schedule — no separate deploy/push step.

Tool	Purpose
`ntro_workflow_create`	Bind a runbook to an entity (with optional schedule)
`ntro_workflow_list` / `_get`	Inspect workflow bindings
`ntro_workflow_run`	Trigger a workflow run — returns a task ID

Agents (external)

Tool	Purpose
`ntro_agent_create`	Register a reference to an external Managed Agent (`anthropic://agents/<id>`, `copilot://agents/<id>`). Provisions a per-agent API key, returns plaintext once.
`ntro_agent_list` / `_get` / `_delete`	Inspect / delete agent registrations
`ntro_agent_create_version`	Bump the agent’s version snapshot (rolling out new prompts)

Tasks (workflow executions)

Tool	Purpose
`ntro_task_get`	Status + step progress for a task
`ntro_task_next_step`	Drive the loop: returns the next action the agent / human should take
`ntro_task_file_ingest`	Submit a file to a task (signal a `submit_file` step)
`ntro_task_data_ingest`	Submit structured data to a task
`ntro_task_approve` / `_reject`	Resolve a review step
`ntro_task_logs`	Tail task execution logs

Schedules

Tool	Purpose
`ntro_schedule_list`	All scheduled / active runs

Resources

Read-only URIs — the MCP client fetches these by URI to populate context.

URI	Description
`ntro://whoami`	Current user + workspace info
`ntro://integrations`	All data platform configs
`ntro://integrations/{id}`	Single data platform detail
`ntro://integrations/{id}/schemas`	Schema discovery (warehouse layout)
`ntro://tenants`	All tenants
`ntro://tenants/{slug}`	Single tenant detail
`ntro://entities`	All entities across tenants
`ntro://workflows`	All workflow definitions
`ntro://workflows/{id}`	Workflow detail
`ntro://tasks/{id}`	Task status + step progress
`ntro://schedule`	All tasks sorted by urgency

Prompts

Guided multi-step flows. The agent receives a structured plan it works through, asking the human for input at key decision points.

Prompt	Description
`onboard_spv`	Walks through SPV setup: entity → COA → email integration → workflow registration → dry-run test
`run_nav`	Monthly NAV execution with progress monitoring and exception handoffs
`platform_status`	Full platform overview — tenants, entities, workflows, active tasks

Self-host (advanced)

Run the MCP server locally if you need to point it at a private Ntropii Workspace, or want to develop against a forked tool surface.

Install

pip (production)
Editable (local dev)

pip install ntro-mcp
ntro-mcp --help

uv pip install -e ~/ntropii/ntro-python --python ~/.ntro-dev/bin/python
uv pip install -e ~/ntropii/ntro-mcp     --python ~/.ntro-dev/bin/python

~/.ntro-dev/bin/ntro-mcp --help

Configuration precedence

The server resolves the API connection in this priority order:

--ntro-host / --ntro-api-key CLI flags
NTRO_HOST / NTRO_API_KEY environment variables
--connection NAME → named entry in ~/.ntro/config.toml
default_connection_name in ~/.ntro/config.toml

Run ntro auth login first (API keys) to seed the config file.

Running locally

stdio (Claude Desktop / Claude Code)
HTTP (claude.ai / other clients)

The MCP client spawns ntro-mcp as a child process and talks over stdin/stdout. Zero networking required.

{
  "mcpServers": {
    "ntro": {
      "command": "/home/you/.ntro-dev/bin/ntro-mcp",
      "args": ["--transport", "stdio", "--connection", "production"]
    }
  }
}

Or use explicit env vars instead of a connection name:

{
  "mcpServers": {
    "ntro": {
      "command": "/home/you/.ntro-dev/bin/ntro-mcp",
      "args": ["--transport", "stdio"],
      "env": {
        "NTRO_HOST": "https://api.ntropii.com/v1",
        "NTRO_API_KEY": "ntro_prod_xxx"
      }
    }
  }
}

Run the server locally and tunnel it with cloudflared for clients that need an HTTPS URL:

# Terminal 1 — start the MCP server
ntro-mcp --transport http --port 8000 --connection production

# Terminal 2 — expose it over HTTPS
cloudflared tunnel --url http://localhost:8000
# → https://random-words.trycloudflare.com

For longer-term use, run on a server you control rather than relying on a development tunnel.

CLI reference

ntro-mcp [OPTIONS]

  --transport {http,stdio}    Transport (default: stdio)
  --port INT                  HTTP port (default: 8000)
  --host TEXT                 HTTP bind host (default: 127.0.0.1)
  -c, --connection NAME       Named connection from ~/.ntro/config.toml
  --ntro-host URL             Workspace API URL (overrides config)
  --ntro-api-key KEY          API key (overrides config)

Coding agents overview

Why MCP matters for design-time runbook authoring.

Claude Code setup

Wire the MCP server into Claude Code.

Copilot Studio setup

Wire the MCP server into Microsoft Copilot Studio.

Ntro CLI

Same surface, exposed for humans in a terminal.

​Quick start — hosted test instance

​Wire it into your client

​What’s included

​Tools

​Identity, tenants, entities

​Integrations (BYO data platforms)

​Runbooks (templates installed on the worker)

​Workflows (runbook → entity bindings)

​Agents (external)

​Tasks (workflow executions)

​Schedules

​Resources

​Prompts

​Self-host (advanced)

​Install

​Configuration precedence

​Running locally

​CLI reference

​Related