# Reported issues for Data Studio

Pod holds 8 of 8 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 [Data Studio](/mcp/data-studio).

## Most discussed

### Epic: MCP Server for Data Studio ecosystem (dockit + sqlkit)

## Epic: MCP Server for Data Studio ecosystem (dockit + sqlkit)

**Status**: Planning
**Target release**: TBD

---

### Background

data-studio-agent powers AI agent loops for both [dockit](https://github.com/geekfun/dockit) (NoSQL: Elasticsearch, OpenSearch, MongoDB, DynamoDB) and [sqlkit](https://github.com/geekfun/sqlkit) (SQL: PostgreSQL, MySQL, SQL Server, SQLite). Currently, only the built-in Tauri-based AI assistant can use these capabilities.

The Model Context Protocol (MCP) has become 

[Read the thread](https://github.com/geek-fun/data-studio-agent/issues/9) · 2026-07-30 · open · 1 comment

### feat: text-only fallback mode for providers without tool support

## Problem

Some LLM providers (Ollama, local models, older endpoints) don't support function calling / tool use. When these models receive tool definitions, they either ignore them or produce malformed responses. The agent loop has no fallback — it either works with tools or fails silently.

## Proposed solution

Add a text_only_fallback option to the agent settings. When enabled:

1. Skip sending tool definitions to the LLM
2. Instead, inject available schema/table/column information into the 

[Read the thread](https://github.com/geek-fun/data-studio-agent/issues/3) · 2026-06-17 · closed · 1 comment

### 发布宣传：提交到 MCP 生态 awesome 列表与聚合平台

## 目标

让 Data Studio Agent 进入主流 MCP 生态的发现渠道，覆盖数据库用户的搜索场景。核心差异化：一个 MCP 服务器通过 DocKit/SqlKit 桥接覆盖 70+ SQL/NoSQL 数据库。

## Phase 1：基础（先做）

- [x] `npm publish --access public`（若未发布）— ✅ 已发布，latest v0.1.10，OIDC provenance
- [x] 添加 GitHub topics：`mcp`、`model-context-protocol`、`database`、`sql`、`nosql`、`doc-kit`、`sql-kit`、`claude`、`cursor`、`vscode` — ✅ 20+ topics 已加（含 claude-code、opencode 等）；`vscode` 未加，可选
- [x] 创建 `server.json`（官方 Registry 格式）并配置 CI 用 GitHub OIDC 发布 — ✅ server.json 已建；release-mcp.yml 已用 `

[Read the thread](https://github.com/geek-fun/data-studio-agent/issues/25) · 2026-08-09 · open · 0 comments

### Bug/UX: MCP server exits with 0 tools when dockit/sqlkit backends are not running — no diagnostics for the agent

## Summary

When the dockit/sqlkit desktop apps are **not running** (or their MCP bridge is not up), `data-studio-mcp` starts up, discovers the stale port files, fails to fetch tools from both backends, and **exits with `process.exit(1)`**. From the agent client's perspective (Claude Code, Codex, Cursor, OpenCode), the MCP server appears as "failed to load / 0 tools" with **no actionable diagnostics** — the LLM cannot tell *why* tools are missing or *how to fix it*.

This is the number-one onboa

[Read the thread](https://github.com/geek-fun/data-studio-agent/issues/21) · 2026-08-07 · closed · 0 comments

### Phase 2: HTTP Bridge — Desktop apps as backends

## [Phase 2] HTTP Bridge — Desktop apps as backends

**Parent epic**: #9
**Status**: Design
**Priority**: High

---

### Background

data-studio-mcp (Phase 3) needs to invoke database operations through dockit and sqlkit, which own all the database drivers, SSH tunnels, and connection management. The bridge is a lightweight HTTP server embedded in each desktop app that exposes the existing `invoke_capability_inner()` and `get_available_tools()` over HTTP.

Only requests from localhost are accept

[Read the thread](https://github.com/geek-fun/data-studio-agent/issues/12) · 2026-07-30 · closed · 0 comments

### Phase 4: MCP Permission Model & Management UI

## [Phase 4] MCP Permission Model — client-driven confirmation + server-enforced policy

**Parent epic**: #9
**Status**: Implemented — all deliverables complete, E2E release-gate pending
**Priority**: High

---

### Background

data-studio-mcp exposes ~63 database tools across ES, MongoDB, DynamoDB (via dockit) and PostgreSQL, MySQL, SQL Server, SQLite (via sqlkit). These tools have different risk levels — reading data vs. creating indexes vs. deleting tables. We need a permission model that:

1

[Read the thread](https://github.com/geek-fun/data-studio-agent/issues/10) · 2026-07-30 · open · 0 comments

### feat: model-aware safety buffer for context compaction

## Problem

The compaction threshold uses a fixed 13,000-token safety buffer regardless of the model's context window. For models with 200K+ windows, this triggers compaction earlier than necessary. For small windows (8K Ollama default), the buffer may be too aggressive.

## Proposed solution

Scale the safety buffer proportionally to the model's context window:

| Context window | Safety buffer |
|---|---|
| ≥ 200K tokens | 8,000 tokens |
| ≥ 32K tokens  | 4,000 tokens |
| < 32K tokens  | 2,000

[Read the thread](https://github.com/geek-fun/data-studio-agent/issues/2) · 2026-06-17 · closed · 0 comments

### feat: parallel execution of read-only tools in agent loop

## Problem

The agent loop currently executes tool calls sequentially — one at a time. When the LLM requests multiple read-only tools (e.g., list_tables + get_columns + get_sample_data), the loop waits for each to complete before starting the next. This adds unnecessary latency for independent operations.

## Proposed solution

Add a parallel_ok: bool field to Capability. When the agent loop receives multiple tool calls, group them into parallel (all parallel_ok) and sequential batches. Execute 

[Read the thread](https://github.com/geek-fun/data-studio-agent/issues/1) · 2026-06-17 · closed · 0 comments

## Most recent

The remaining reports are on [the project's issue tracker](https://github.com/geek-fun/data-studio-agent/issues).