[Proposal] Nacos Plugin for Coding Agents — Bring Nacos capabilities to Claude Code, Cursor, and beyond

### Is your feature request related to a problem? Please describe.

Nacos 3.2.0 completed the **AI Registry Suite** on the server side — MCP Registry, Agent Registry, Skill Registry, Prompt Registry, and Copilot. The server-side capabilities are comprehensive and production-ready.

However, **these capabilities don't reach developers where they actually work**. A new category of developer tooling — **Coding Agents** (Claude Code, Cursor, Windsurf, Gemini CLI, GitHub Copilot) — is rapidly becoming the primary interface for software development. Developers write code, debug, manage infrastructure, and interact with services through these agents without leaving the terminal or editor.

Today, when a developer working inside a coding agent wants to:
- Check a Nacos configuration → open browser, navigate to Nacos Console, find the config
- Look up registered services → switch context to Console or manually construct curl commands
- Sync team AI Skills → run nacos-cli commands with the right parameters from memory

This context-switching breaks the developer flow. The server has the capabilities, but the coding agents can't access them.

**What if a developer could just say:** _"Check the Redis config for order-service in staging"_ — and get the answer instantly, right where they're working?

Nacos 3.2.0 在 Server 端完成了完整的 AI 注册中心能力（MCP Registry、Agent Registry、Skill Registry、Prompt Registry、Copilot），但这些能力无法直接触达开发者日常工作的 Coding Agent 环境（Claude Code、Cursor、Gemini CLI 等）。开发者需要频繁切换上下文到 Nacos Console 才能查配置、查服务、同步 Skill，这打断了开发流程。

---

### Describe the solution you'd like

A **Nacos Plugin** for coding agents that teaches AI agents to interact with Nacos Server directly via REST API.

The key insight: modern coding agents (Claude Code, Cursor, etc.) can execute shell commands including `curl`. A plugin doesn't need to ship any binary — it just needs to provide **Markdown instructions** (Skills, Commands, Agents) that teach the AI agent which APIs to call and how to present the results.

#### How it works

```
Developer: "Check the Redis config for order-service in staging"
     ↓
Coding Agent reads plugin's Skill → learns the API workflow
     ↓
Executes: curl POST /nacos/v3/auth/user/login → gets accessToken
     ↓
Executes: curl GET /nacos/v3/console/cs/config?dataId=redis.yaml&groupName=order-service&namespaceId=staging
     ↓
Coding Agent formats and presents the config to developer
```

No binary installation. No SDK. Just Markdown files that encode Nacos API knowledge.

核心思路：现代 Coding Agent 可以执行 shell 命令（包括 `curl`）。Plugin 不需要发布任何二进制文件——只需提供 **Markdown 指令**（Skills、Commands、Agents），教 AI Agent 调用哪些 API、如何解析结果。零依赖，装完即用。

#### Plugin structure

```
nacos-plugin/
├── .claude-plugin/plugin.json       # Plugin metadata
├── commands/                        # Slash commands (one-shot operations)
│   ├── nacos-config.md              # /nacos-config — query/manage configs
│   ├── nacos-service.md             # /nacos-service — query services & instances
│   └── nacos-skill.md               # /nacos-skill — query/sync AI Skills
├── agents/                          # Specialized sub-agents
│   └── nacos-explorer.md            # Deep exploration of Nacos data
├── skills/                          # Multi-step workflows
│   ├── manage-config/SKILL.md       # Configuration management workflow
│   ├── query-service/SKILL.md       # Service discovery & health check
│   ├── sync-skills/SKILL.md         # Team Skill synchronization
│   └── publish-skill/SKILL.md       # Publish Skill to Nacos
└── references/
    └── nacos-api-reference.md       # Nacos v3 API reference for agents
```

Three capability layers:

| Layer | Purpose | Example |
|-------|---------|---------|
| **Commands** | One-shot queries | `/nacos-config redis.yaml` → instant result |
| **Skills** | Multi-step workflows | "Sync all team Skills to my local env" |
| **Agents** | Deep analysis | "Find what config change caused the production error" |

#### Concrete example: what a Command looks like

Here's a simplified `commands/nacos-config.md` showing how Markdown teaches an AI agent to call Nacos APIs:

```markdown
---
description: "Query or manage Nacos configuration. Usage: /nacos-config <dataId> [group] [namespace]"
argument-hint: "<dataId> [group] [namespace]"
allowed-tools: [Bash, Read]
---

## Steps

1. **Authenticate** (if auth enabled):
   POST http://${NACOS_ADDR}/nacos/v3/auth/user/login
   Body: username=${NACOS_USERNAME}&password=${NACOS_PASSWORD}
   → Extract accessToken from response

2. **Fetch config**:
   GET http://${NACOS_ADDR}/nacos/v3/console/cs/config
     ?dataId=<dataId>&groupName=<group>&namespaceId=<namespace>
   Header: Authorization: Bearer <token>

3. **Present** the config content with syntax highlighting based on file type.

On 403 → re-authenticate. On 404 → suggest listing configs.
```

**This Markdown file IS the entire implementation.** No code compilation, no binary, no deployment.

#### Use cases

**1. Configuration Query**
```
Developer: "Show me the database config for payment-service in production"
Agent: Authenticates → GET /nacos/v3/console/cs/config → presents formatted YAML
```

**2. Service Discovery**
```
Developer: "How many instances of order-service are healthy?"
Agent: GET /nacos/v3/console/ns/service/list → GET /nacos/v3/console/ns/instance/list
     → presents table: IP, port, health status, metadata
```

**3. Config Change Investigation**
```
Developer: "Production order-service started failing yesterday. Check recent config changes."
Agent (nacos-explorer): Lists configs → GET /nacos/v3/console/cs/history/list for each
     → filters last 24h → GET /nacos/v3/console/cs/history?nid=<id> for details
     → "Found: connection pool size reduced 50→20 at 14:30, Redis timeout 3000→500ms at 16:00"
```

**4. Team Skill Sync**
```
Developer: "Sync the latest team AI Skills to my local environment"
Agent: GET /nacos/v3/admin/ai/skills/list → fetches each skill
     → reconstructs SKILL.md + resources → writes to local directory
     → "Synced 12 skills: 3 updated, 9 unchanged"
```

#### Progressive enhancement

The plugin evolves in phases without breaking changes:

| Phase | Approach | Dependency | Coverage |
|-------|----------|------------|----------|
| **A** | Skills teach agents to use `curl` | **None** | Claude Code, Cursor |
| **B** | `nacos-cli mcp serve` (stdio MCP Server) | nacos-cli | + Windsurf, Gemini CLI, Copilot |
| **C** | Nacos Server native Remote MCP endpoint | **None** (server built-in) | All MCP clients |

Phase A is the starting point — zero dependency, immediate value. Phase B adds MCP tools for richer integration across more coding agents. Phase C is the long-term vision when Nacos Server exposes a native MCP endpoint.

Plugin 分三阶段演进：Phase A 零依赖（纯 curl），Phase B 加 MCP Server（需 nacos-cli），Phase C 远程 MCP（Server 原生支持后）。每个阶段向后兼容。

---

### Why this matters for Nacos

**1. Marketplace gap**

The Claude Code Plugin Marketplace currently has **61 plugins** (Development, Productivity, Database, Monitoring, Security, etc.). **Zero plugins exist for infrastructure management, configuration centers, or service discovery.** Nacos would be the first — defining a new category.

Claude Code Plugin Marketplace 目前有 61 个插件，覆盖多个类别，但**基础设施管理、配置中心、服务发现类为零**。Nacos 将开辟全新品类。

**2. Closing the supply-demand loop**

Nacos 3.2.0 built the supply side (AI Registry on Server). This plugin builds the demand side (consumption in coding agents). Together:

```
Team Lead crafts Skills → Nacos Skill Registry (Server)
     ↓
Developer installs Plugin → Coding Agent calls Nacos API to fetch Skills
     ↓
Developer uses Skills in daily workflow → creates new Skills → publishes back
```

The Skill Registry (#14110) becomes truly useful when Skills flow seamlessly between Nacos and the coding agents that consume them.

**3. Multi-agent coverage from a single implementation**

| Coding Agent | Phase A | Phase B (MCP) |
|-------------|---------|---------------|
| Claude Code | ✅ | ✅ |
| Cursor | ✅ | ✅ |
| Windsurf | — | ✅ |
| Gemini CLI | — | ✅ |
| GitHub Copilot | — | ✅ |

**4. New audience**

The plugin puts Nacos in front of the AI-native developer community — a different and rapidly growing audience beyond the traditional Java/microservices ecosystem.

---

### Relationship to existing work

| Component | Owner | This Proposal |
|-----------|-------|---------------|
| Config / Naming / Skill API | Server team | **Consumer** — calls these APIs |
| MCP Registry (`/v0/servers`) | Server team | Complementary — different audience |
| Copilot | Server team | Non-overlapping — Copilot in Console, Plugin in coding agents |
| nacos-cli | CLI team | Phase B enhancement — adds `mcp serve` |

**Phase A does NOT modify any Server or CLI code.** It is pure Markdown. Phase B adds one command to nacos-cli.

---

### Describe alternatives you've considered

1. **Improve nacos-cli only** — Developers can already use nacos-cli commands. But this requires installation, remembering command syntax, and doesn't integrate with the coding agent's context. The plugin approach is zero-dependency and context-aware.

2. **Wait for Nacos Server to natively expose MCP endpoints** — This would be ideal long-term (Phase C), but requires significant server-side work. Phase A delivers value immediately with zero server changes.

3. **Build a traditional IDE extension** (VSCode/IntelliJ) — Different audience. Coding agents are a new paradigm that complements but doesn't replace traditional IDE extensions.

---

### Additional context

- **Submission**: The plugin would be submitted to the official marketplace via https://clau.de/plugin-directory-submission
- **Repository**: Suggested to live in `nacos-group/nacos-cli` (co-located, like Atlassian's approach in `atlassian/atlassian-mcp-server`)
- **Auth**: Environment variables (`NACOS_ADDR`, `NACOS_USERNAME`, `NACOS_PASSWORD`) — the standard approach in plugin ecosystems

#### Nacos v3 API endpoints used

| Operation | Method | Path |
|-----------|--------|------|
| Login | POST | `/nacos/v3/auth/user/login` |
| Get config | GET | `/nacos/v3/console/cs/config` |
| List configs | GET | `/nacos/v3/console/cs/config/list` |
| Config history | GET | `/nacos/v3/console/cs/history/list` |
| List services | GET | `/nacos/v3/console/ns/service/list` |
| List instances | GET | `/nacos/v3/console/ns/instance/list` |
| List skills | GET | `/nacos/v3/admin/ai/skills/list` |
| Get skill | GET | `/nacos/v3/admin/ai/skills` |

All paths include the default `/nacos` context path. Full parameter details will be in `references/nacos-api-reference.md`.

---

I'd love to hear from the community — especially developers who use both Nacos and coding agents in their daily workflow. Which use cases would be most valuable to you?

期待社区反馈，特别是同时使用 Nacos 和 Coding Agent 的开发者——哪些场景对你最有价值？

*Related: #14110 (Skills Registry Discussion)*

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Proposal] Nacos Plugin for Coding Agents — Bring Nacos capabilities to Claude Code, Cursor, and beyond #14603

Is your feature request related to a problem? Please describe.

Describe the solution you'd like

How it works

Plugin structure

Concrete example: what a Command looks like

Use cases

Progressive enhancement

Why this matters for Nacos

Relationship to existing work

Describe alternatives you've considered

Additional context

Nacos v3 API endpoints used

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Layer	Purpose	Example
Commands	One-shot queries	`/nacos-config redis.yaml` → instant result
Skills	Multi-step workflows	"Sync all team Skills to my local env"
Agents	Deep analysis	"Find what config change caused the production error"

Phase	Approach	Dependency	Coverage
A	Skills teach agents to use `curl`	None	Claude Code, Cursor
B	`nacos-cli mcp serve` (stdio MCP Server)	nacos-cli	+ Windsurf, Gemini CLI, Copilot
C	Nacos Server native Remote MCP endpoint	None (server built-in)	All MCP clients

Coding Agent	Phase A	Phase B (MCP)
Claude Code	✅	✅
Cursor	✅	✅
Windsurf	—	✅
Gemini CLI	—	✅
GitHub Copilot	—	✅

Component	Owner	This Proposal
Config / Naming / Skill API	Server team	Consumer — calls these APIs
MCP Registry (`/v0/servers`)	Server team	Complementary — different audience
Copilot	Server team	Non-overlapping — Copilot in Console, Plugin in coding agents
nacos-cli	CLI team	Phase B enhancement — adds `mcp serve`

Operation	Method	Path
Login	POST	`/nacos/v3/auth/user/login`
Get config	GET	`/nacos/v3/console/cs/config`
List configs	GET	`/nacos/v3/console/cs/config/list`
Config history	GET	`/nacos/v3/console/cs/history/list`
List services	GET	`/nacos/v3/console/ns/service/list`
List instances	GET	`/nacos/v3/console/ns/instance/list`
List skills	GET	`/nacos/v3/admin/ai/skills/list`
Get skill	GET	`/nacos/v3/admin/ai/skills`

[Proposal] Nacos Plugin for Coding Agents — Bring Nacos capabilities to Claude Code, Cursor, and beyond #14603

Description

Is your feature request related to a problem? Please describe.

Describe the solution you'd like

How it works

Plugin structure

Concrete example: what a Command looks like

Use cases

Progressive enhancement

Why this matters for Nacos

Relationship to existing work

Describe alternatives you've considered

Additional context

Nacos v3 API endpoints used

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions