# MooseStack is EOL: hypequery vs Moose for ClickHouse (https://hypequery.com/compare/hypequery-vs-moose)

Updated: 2026-11-08

**Verdict:** MooseStack has reached end of life and is no longer actively maintained. Do not start a new production system on it. hypequery is the active, narrower replacement for its query models, APIs, React hooks, and MCP tools, in TypeScript or Python.

| | hypequery | MooseStack |
|---|---|---|
| Project status | Adding typed queries and APIs to a ClickHouse you already run | End of life and no longer actively maintained |
| Schema source | Introspected from your live ClickHouse schema | Defined in code and migrated into ClickHouse |
| Migration boundary | Typed queries, semantic datasets, APIs, React hooks, and MCP | Move DDL, Redpanda, and Temporal responsibilities separately |
| Scope | Actively maintained TypeScript and Python packages inside your application | Unmaintained framework code can inform migration but should not be a new dependency |
| Languages | TypeScript; Python SDK in beta | TypeScript and Python |
| Streaming and workflows | Not covered: keep dedicated ingestion and workflow tools | Redpanda and Temporal modules, now unmaintained |
| Agents and MCP | Governed dataset and metric MCP server, from either SDK | Development harness MCP and query-model tools |

## When to keep MooseStack for now

- Your Moose app runs in production and works today: migrate one query model at a time rather than in one cutover.
- You depend on Moose’s streaming and workflow modules and have not chosen replacements yet. hypequery does not cover streaming, workflows or DDL migrations.
- Your ClickHouse schema is declared in Moose code and no migration tool owns the DDL yet.

## MooseStack has reached end of life

The decision is no longer between two actively maintained projects. In June 2026 the MooseStack maintainers announced that **MooseStack has reached end of life and is no longer actively maintained**. Read the [official MooseStack EOL statement on GitHub](https://github.com/514-labs/moosestack#readme).

That means we do not recommend starting a new production analytics backend on MooseStack. Existing users should first decide which parts of Moose they actually need to replace: ClickHouse schema management, streaming ingestion, workflows, query models and APIs, or the agent-facing development harness. hypequery covers the query model, API, React and MCP portion, in TypeScript or Python. It does not attempt to replace Redpanda, Temporal, or a migration system.

For a step-by-step inventory of a TypeScript project, use the [MooseStack to hypequery migration guide](/blog/migrating-moosestack-to-hypequery). For a Python project, start with the [Python SDK docs](/docs/python).

## How Moose maps to hypequery

Moose shipped TypeScript and Python libraries, and so does hypequery. The Python SDK is in beta.

| Moose | hypequery TypeScript | hypequery Python (beta) |
| --- | --- | --- |
| `OlapTable` data models | Types generated from the live schema with `npx hypequery generate` | Dataset scaffolds generated from the live schema with `hypequery generate datasets` |
| Query models: `defineQueryModel()` / `define_query_model()` | Datasets with dimensions, measures, filters and relationships (`@hypequery/datasets`) | The same dataset model in the `hypequery` package |
| Agent tools: `registerModelTools()` / `create_model_tool()` | `@hypequery/mcp` exposes datasets and metrics as MCP tools | `hypequery mcp` serves datasets over stdio |
| Query APIs (`Api`) | `@hypequery/serve`: validated routes with OpenAPI, mounted in Next.js, Node.js or any Fetch runtime | Serve on FastAPI, Django or Flask (ASGI, WSGI, or native URLs and blueprints) |
| Dashboards | Typed React hooks on the same contract | The same React hooks, generated from the Python API's exported client contract |
| Tenant scoping | Tenant scope enforced from trusted runtime context | The same, including your app's own login (`django_user`, `flask_session`) |
| Streaming, workflows, DDL migrations | Not covered: bring dedicated tools | Not covered: bring dedicated tools |

TypeScript also includes a ClickHouse-native query builder for queries that stay local to one service. Python has no query builder; queries go through datasets. Python adds typed DataFrames (`to_pandas`, `to_polars`, `to_arrow`) and runs on embedded ClickHouse through chDB for local development and tests. The [Python feature status](/docs/python/feature-status) page tracks what is available today.

## What hypequery replaces

If your Moose application used typed data models, query models and APIs to power a product, hypequery provides the active, narrower path:

- generate types or dataset scaffolds from the ClickHouse schema that exists now;
- move reusable analytics into typed datasets, dimensions, measures and metrics;
- expose those contracts through validated HTTP routes;
- consume them through typed React hooks;
- expose governed metrics and dataset queries through MCP tools for AI agents;
- enforce multi-tenant scope from trusted runtime context.

The direction of schema ownership changes. Moose was code-first infrastructure: code declared tables and pushed schema into ClickHouse. hypequery is database-first at the physical layer and code-first at the semantic layer: your migration tool owns ClickHouse DDL, the hypequery CLI introspects the live result, and your TypeScript or Python code owns the product-facing analytics contract.

## What hypequery does not replace

hypequery is not a streaming or workflow platform. Keep or choose dedicated tools for:

- Kafka or Redpanda ingestion;
- Temporal workflows and scheduled orchestration;
- ClickHouse DDL migrations;
- infrastructure provisioning and deployment of ClickHouse itself.

This smaller boundary is deliberate. It lets teams migrate the application-facing analytics layer without rebuilding ingestion and infrastructure at the same time.

## Current comparison

| Area | MooseStack at EOL | hypequery |
| --- | --- | --- |
| Maintenance | End of life; no longer actively maintained | Actively developed open-source packages |
| Languages | TypeScript and Python | TypeScript; Python in beta |
| Physical schema | Declared in Moose code and migrated into ClickHouse | Introspected from the live ClickHouse schema |
| Semantic model | Query models over data models | Datasets with dimensions, measures, metrics and relationships, in TypeScript or Python |
| Query builder | Moose query layer | ClickHouse-native typed builder in TypeScript, with FINAL, LIMIT BY, percentiles, argMax, CTEs and expressions |
| HTTP APIs | Framework-managed query APIs | Validated Serve routes inside your app: Node.js, Next.js and Fetch runtimes in TypeScript; FastAPI, Django and Flask in Python |
| React | Application-owned integration | Typed TanStack Query hooks, from either SDK |
| Agents | Development harness MCP and query-model tools | Governed dataset and metric MCP server, from either SDK |
| Streaming and workflows | Redpanda and Temporal modules | Bring your existing tools |

## Migration shape

The safest migration is incremental:

1. Freeze new Moose-specific surface and export the current ClickHouse DDL, tables and materialized views.
2. Assign schema migrations, ingestion and workflows to explicit tools outside the application query layer.
3. Generate hypequery types (TypeScript) or dataset scaffolds (Python) from the live ClickHouse schema.
4. Port one query model at a time to a dataset and compare the generated SQL and results.
5. Promote shared calculations into measures and metrics.
6. Move API consumers to Serve routes, then React hooks or MCP tools.
7. Remove the Moose runtime only after traffic, jobs and deployment dependencies have been accounted for.

Both can run against the same ClickHouse during the transition, so there is no reason to combine the infrastructure and query migrations into one cutover.

## Recommendation

For a new application on ClickHouse, use an actively maintained semantic layer rather than adopting an unmaintained framework. hypequery is the fit when you already have, or want to keep control of, ClickHouse and need typed datasets, multi-tenant analytics APIs, React hooks and MCP tools, from TypeScript or Python.

If you need the full infrastructure scope Moose once provided, pair hypequery with the dedicated ingestion, workflow and migration tools your team is prepared to operate. Start with the [migration guide](/blog/migrating-moosestack-to-hypequery) for TypeScript or the [Python quick start](/docs/quick-start#python), check the [current capability matrix](/docs/capabilities), and test it against one real table.

## FAQ

### Are hypequery and Moose solving the same problem?

They overlap on typed ClickHouse queries and APIs, but Moose also owned schema, streaming, workflows, and a development runtime. hypequery replaces the application-facing analytics layer; dedicated tools should own the other responsibilities.

### Should I choose MooseStack for a new project?

No. MooseStack’s maintainers announced end of life and the project is no longer actively maintained. Use the historical architecture as migration context, not as a new production dependency.

### I used MooseStack from Python. Is there a Python path?

Yes. The hypequery Python SDK (beta) defines the same datasets in Python, serves them from FastAPI, Django, or Flask, exposes them to agents with hypequery mcp, and feeds the same React hooks through an exported client contract. Python has no query builder; queries go through datasets.

### Can hypequery replace every MooseStack module?

No. hypequery covers typed ClickHouse queries, semantic datasets, APIs, React hooks, multi-tenancy, and MCP. Keep or choose separate tools for DDL migrations, streaming ingestion, and workflow orchestration.
