# Reported issues for Mako

Pod holds 7 of 7 problems reported by people outside the maintainer team. Issues filed by the project's own owners, members and collaborators are excluded entirely — a maintainer's release checklist is not a warning to a prospective user.

Back to [Mako](/mcp/mako).

## Most discussed

### MCP notebook bridge: add_notebook_cell / edit_notebook_cell report success but cell source is never persisted (silent write failure)

## Summary

Over the MCP surface (`POST /api/mcp`, workspace API key), `add_notebook_cell` and `edit_notebook_cell` return `success: true` while the cell **source is never persisted**. Reading the cell back shows empty source, and "running" the cell executes the empty source and reports ok. Nothing in the loop ever surfaces an error.

Silent success on a failed write is the worst possible failure mode for an agent bridge: the calling agent (and its user) walk away believing a notebook exists wit

[Read the thread](https://github.com/mako-ai/mako/issues/764) · 2026-07-30 · open · 1 comment

### RFC: App Data — persistent CRUD storage for Mako apps

# RFC: App Data — a provisioned serverless database per app

> **v2** — reworked after design discussion. v1 proposed a records store in Mako's MongoDB; we decided against putting app data in Mongo. App data now lives in a **provisioned serverless SQL database**, registered as a regular Mako connection, with the storage backend **configurable** (Mako-managed by default, BYO later).

## Problem

Mako apps are read-only data products. They can query workspace data through bindings, but they have n

[Read the thread](https://github.com/mako-ai/mako/issues/763) · 2026-07-30 · open · 0 comments

### Server-authoritative console drafts + live push: realtime sync for multiplayer, multi-tab, and detached agents

Follow-up to #467 / PR #472. The resumable-stream foundation (server-owned generation, reattach on refresh, turn-start persistence, console save version guard) is in place. The remaining problems below all share one root cause and are solved by one architectural change, described after them.

---

## The problems

### P1 — Stream breaks on client tool calls when the client is detached
If the agent issues a client-side tool call (`modify_console`, `read_console`, …) and no client is attached (ref

[Read the thread](https://github.com/mako-ai/mako/issues/475) · 2026-06-10 · open · 0 comments

### Resumable chat streams: decouple generation from client connection (server owns the stream)

## Goal

Make chat/agent generation survive client disconnects and support resumable streams + parallel agents, matching the ChatGPT/Cursor architecture: **the server owns the stream, the browser is just a reader.**

Companion issue on the Aura side (same architecture discussion, 2026-06-09): https://github.com/AuraHQ-ai/aura/issues/1111. Aura is Vercel-hosted, so it gets WDK. Mako's constraints are different, so the recommendation differs — analysis below.

## Current state (verified against th

[Read the thread](https://github.com/mako-ai/mako/issues/467) · 2026-06-09 · open · 0 comments

### RFC: User-Defined Sandboxed Connectors — Vibe-Coded Data Integrations

# PRD: User-Defined Sandboxed Connectors

> Vibe-coded data integrations for Mako

## Summary

Let users build, test, and deploy connectors for any third-party API from within Mako's UI. Connector source code lives in the database, is editable on the fly with AI assistance, and executes in sandboxed E2B environments. Users write TypeScript, import any npm package, and Mako handles scheduling, retries, schema management, and writes to any connected database.

---

## Motivation

### Current State

[Read the thread](https://github.com/mako-ai/mako/issues/140) · 2026-03-19 · open · 0 comments

### Smart query execution -- hybrid timeout + client-side run

## Part of Epic: #108

## Problem

The universal prompt has a "Test Before Deliver" directive: the agent must execute a query successfully before writing it to the console via `modify_console`. This prevents broken SQL from reaching the user -- but has a critical flaw:

1. Agent executes query (server-side) -- may timeout on large/slow queries
2. Agent writes to console
3. User clicks "Run" -- executes the same query again
4. User waits **twice** for the same result

Worse: if step 1 times out, 

[Read the thread](https://github.com/mako-ai/mako/issues/110) · 2026-03-05 · closed · 0 comments

### feat: tier-gate premium AI models behind Pro plan (stop token bleeding)

## Problem

All AI models (including Claude Opus 4.6 — the most expensive) are available to every user with zero billing checks. Free-tier workspaces can burn unlimited tokens on flagship models. Currently hundreds of users are chatting with Opus 4.6 non-stop.

## Solution: Tier-gated model access

Gate premium models behind `pro`/`enterprise` billing tiers. Free-tier users see premium models in the selector but can't use them — they get a "Book a Demo" upgrade prompt instead. No Stripe integrat

[Read the thread](https://github.com/mako-ai/mako/issues/98) · 2026-02-24 · open · 0 comments

## Most recent

The remaining reports are on [the project's issue tracker](https://github.com/mako-ai/mako/issues).