Use it with AI. MCP, in plain words.

"MCP" sounds technical, but the idea is simple: it's a shared way for AI assistants — Claude, an agent IDE, or one you build — to safely use a platform's tools and data. CoTrackPro publishes its catalog as MCP tools, so an assistant can look things up and draft documents under the same safety rules as everything else here.

Educational and informational only — not legal or clinical advice. Every AI-generated draft is a starting point for a person to review — never a substitute for a licensed professional.

What an "MCP server" is#

An MCP server publishes a set of tools (things an assistant can do, like "search the artifact catalog" or "draft an incident log") and resources (things it can read).
An AI host — Claude, the Antigravity IDE, your own agent — connects to the server and calls those tools when it needs them. The host speaks the protocol; you don't wire it by hand.
The CoTrackPro MCP server exposes the platform's catalog as tools: discover roles, workflows, and scenarios; look up any of the 133 artifact schemas; and generate court-neutral, child-centered drafts — always crisis-first, always educational-only.

It's the same catalog you can already browse here — MCP is just the live way an AI assistant reaches it.

The tools the server publishes#

These are the actual tools the CoTrackPro MCP server exposes — the live counterpart to the offline catalog. Discovery and read tools are free; generation tools are educational-only and crisis-first. Browse the same data offline with cotrackpro catalog.

Group	Tools	What they do
Get oriented	`get_started`, `find_tool_for`, `find_example_for`	Persona picker (crisis-first), route an intent to the right tool, jump to a worked example.
Discover the catalog	`list_roles`, `list_workflows`, `list_skills`, `list_artifacts`, `list_scenarios`, `list_prompt_modules`	Enumerate the typed catalog — the live source the offline SDK mirror is generated from.
Inspect one thing	`get_role`, `get_workflow`, `get_artifact_schema`, `get_scenario`, `get_example_case`	A role's composed skills + guardrails, a workflow's full step graph, an artifact's JSON Schema, a worked example.
Search	`search_artifacts`	Rank the 133 artifacts by keyword.
Compose a diagram	`get_role_diagrams`, `get_role_diagrams_rendered`, `get_workspace_map`	Per-role Mermaid diagrams over the role → trigger → workstream → artifact graph.
Draft (educational only)	`draft_artifact`, `bulk_draft_artifacts`, `rewrite_coparent_message`, `start_from_example`	Generate a first draft under child-centered, trauma-informed, court-neutral guardrails.

Every role and skill the server returns carries its child-centered / trauma-informed guardrail modules — the same lens the catalog is organized around. Run cotrackpro catalog safety to see which roles, skills, and modules carry each lens offline.

New to terms like terminal, Node.js, or npm? The Developer tools page explains each one in plain words.

Three ways to use it#

🔎

Just explore it — no setup#

The CLI and SDK ship the whole catalog typed and offline, so you can see exactly what the MCP server offers without connecting anything.

🤖

In your AI agent — typed#

Your AI host already speaks MCP. The SDK gives you a typed wrapper so you call platform tools by name instead of hand-writing requests.

🧩

Local MCP tools via the CLI#

Install and run MCP-server plugins through the CLI. It brokers a short-lived, scoped token on launch, so the tool never sees your credentials.

# 1. Explore the catalog offline — the fastest way to understand what's there:
cotrackpro catalog                       # counts across every kind
cotrackpro catalog roles                 # roles | workflows | skills | knowledge | scenarios | prompt-modules
cotrackpro catalog safety                # the child-centered & trauma-informed lens
cotrackpro artifacts search incident     # find an artifact by keyword

# 3. Run a local MCP tool through the CLI broker (no credentials in the tool):
cotrackpro plugins register .            # register a local MCP plugin
cotrackpro mcp launch my-tool            # the CLI injects a scoped token on launch

For #2, the SDK is a thin typed binding over whatever MCP plumbing your host already has — it does not re-implement the protocol:

import { CotrackproToolset } from "@cotrackpro/sdk/mcp";

// `host.callTool` is your AI host's MCP call (Claude runtime, Antigravity, an agent SDK).
const cp = new CotrackproToolset((name, args) => host.callTool(`cotrackpro.${name}`, args));

const artifacts = await cp.listArtifacts();          // typed results
const schema = await cp.getArtifactSchema("incident-log");

Connect it to your AI app — copy & paste#

Most AI apps let you add an MCP server by pasting a small config. The public CoTrackPro MCP server is https://mcp.cotrackpro.com — copy the one for your app and restart it. (Running a private deployment? Swap in your own URL.)

Claude Desktop — add to claude_desktop_config.json:

{
  "mcpServers": {
    "cotrackpro": {
      "url": "https://mcp.cotrackpro.com"
    }
  }
}

Claude Code (terminal):

claude mcp add --transport http cotrackpro https://mcp.cotrackpro.com

Cursor — add to .cursor/mcp.json:

{
  "mcpServers": {
    "cotrackpro": {
      "url": "https://mcp.cotrackpro.com"
    }
  }
}

Prefer the terminal? The CLI prints the exact snippet for your app (it defaults to the public server; pass --url for a private one):

cotrackpro mcp config --client cursor

Once it's connected, your assistant can use CoTrackPro's tools directly — try asking it to "use CoTrackPro to get started" (it calls get_started), and it'll walk you through, crisis-resources first. Not sure where to begin? Paste this prompt:

I'm new to CoTrackPro. Using its tools, show me the persona picker, then give me a
short, plain-language quickstart for documenting a co-parenting situation — and lead
with crisis resources. Keep everything educational, not legal or clinical advice.

How it all fits#

CoTrackPro MCP server   →   @cotrackpro/sdk/mcp registries   →   the CLI & this portal
   (live tools an AI         (a typed, OFFLINE mirror —            (read the mirror; work
    host calls)               your code can't drift, and            without a connection)
                              works with no network)

The MCP server is the live tool surface; the SDK registries are a typed, offline mirror of it, generated from the catalog and drift-gated — so your code stays in sync and still runs with no connection. The CLI and this site read that mirror. Same catalog, reached three ways.

Machine-readable catalog & discovery#

Building your own integration? The catalog is published as a stable, versioned contract you can fetch directly — no private repo, no auth, no SDK:

Discovery: /.well-known/mcp.json — the server endpoint, the catalog URL, connect instructions, the documented tool surface, and the safety contract, in one fetch.
Catalog contract: /data/catalog.v1.json — every artifact, role, workflow, skill, knowledge file, scenario, and prompt module, plus the child-centered / trauma-informed taxonomy. Versioned (1.0) and drift-gated against the code.
JSON Schema: /data/catalog.schema.json — validate before you build.

Same catalog the CLI and the live server use — published as a static contract anyone can build on.

A few words, in plain terms#

MCP (Model Context Protocol) — a shared standard for letting AI assistants use tools and data.
MCP server — a program that publishes those tools (here: the CoTrackPro catalog).
Tool / resource — something an assistant can do (a tool) or read (a resource).
Host — the AI app that connects to the server and calls its tools (Claude, an agent IDE, your code).

Safety, always#

Every generation tool is educational and informational only — not legal or clinical advice. The server leads with crisis resources, keeps language court-neutral and child-centered, and never places confidential family-case information into a shared space. If you or a family is in crisis, reach out first: 988, text HOME to 741741, 1-800-799-7233, 1-800-422-4453, or 911.