feat: add public feature request chat app

[viktormarinho]

Summary

Adds a new Next.js app at apps/feature-request/ — a public-facing chat page where users can propose features for MCP Mesh. An AI "tech lead" persona reads the codebase via MCP tools, asks clarifying product/UX questions, builds a structured plan, and creates a GitHub issue.

• Zero changes to apps/mesh/ — this is a standalone consumer of Mesh's existing decopilot streaming API, authenticated via API key
• 15 new files, 812 lines — Next.js 15, AI SDK v3, Tailwind v4, @deco/ui design system
• Deploys to Vercel with zero config (or any platform that runs Next.js)

How it works

Browser → POST /api/chat → Next.js API route → Mesh decopilot/stream (Bearer API key)
                                ↓
                         Virtual MCP agent
                                ↓
                    GitHub MCP tools (read code, create issues)

1. User describes a feature idea in the chat
2. The AI silently reads the codebase via GitHub MCP tools, asks product-oriented clarifying questions
3. Builds a two-layer plan: user-facing summary + technical implementation notes
4. Creates a GitHub issue with the full plan when the user approves

Key files

File	Purpose
app/api/chat/route.ts	Proxy to Mesh — injects API key, system prompt, model/agent config
app/page.tsx	Chat page with empty state, ice breakers, message list, input
lib/system-prompt.ts	Tech lead persona: translates code into plain language, asks UX questions
middleware.ts	IP-based rate limiting (5 req/min) to prevent abuse
components/chat-messages.tsx	Message rendering with friendly tool call indicators ("Reading file...", "Creating issue...")

Configuration

All config via env vars (.env.local for dev, Vercel dashboard for prod):

MESH_URL=http://localhost:3000
MESH_API_KEY=mk_xxx
MESH_ORG_SLUG=my-org
AGENT_ID=vir_xxx           # Virtual MCP with GitHub MCP connected
MODEL_CONNECTION_ID=conn_xxx
MODEL_ID=google/gemini-2.5-flash
TOOL_MODE=smart_tool_selection

Required MCP tools (GitHub MCP server)

The virtual MCP agent needs a GitHub MCP connection with toolsets: repos,issues,context,git — specifically: get_file_contents, search_code, get_repository_tree, list_commits, search_issues, list_issues, issue_write, get_me.

Test plan

• Set up a virtual MCP agent in Mesh with GitHub MCP connection
• Create an API key in Mesh, configure .env.local
• Run bun run dev (Mesh on :3000) and cd apps/feature-request && bun run dev (on :3003)
• Open localhost:3003, describe a feature, verify the AI reads code and asks clarifying questions
• Verify issue creation flow works end-to-end
• Verify rate limiting returns 429 after 5 rapid requests
• Run next build to confirm production build succeeds

Made with Cursor

---

Summary by cubic

Add a standalone Next.js app at apps/feature-request for public feature requests via chat; a virtual tech lead reads the codebase with MCP tools, asks clarifying questions, drafts a plan, and can create a GitHub issue. It proxies Mesh’s decopilot stream with an API key and does not change apps/mesh.

• New Features

  • Chat UI with empty state, ice breakers, message list, and input
  • Tool-call indicators (e.g., Reading file, Creating issue)
  • Next.js API route that injects the system prompt and streams from Mesh
  • IP-based rate limiting (5 req/min)
  • Env-based config; deploys on Vercel without extra setup

• Migration

  • Set env vars: MESH_URL, MESH_API_KEY, MESH_ORG_SLUG, AGENT_ID, MODEL_CONNECTION_ID, MODEL_ID, TOOL_MODE
  • Ensure the virtual MCP agent has GitHub tools: repos, issues, context, git
  • Run the app at apps/feature-request and verify end-to-end issue creation

^{Written for commit 7416268. Summary will update on new commits.}

[cubic-dev-ai]

1 issue found across 15 files

Prompt for AI agents (all issues)

Check if these issues are valid — if so, understand the root cause of each and fix them.


<file name="apps/feature-request/app/api/chat/route.ts">

<violation number="1" location="apps/feature-request/app/api/chat/route.ts:43">
P2: Validate that `messages` is an array before spreading it. As written, a malformed or missing `messages` field will throw and return a 500 instead of a client error.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}

[cubic-dev-ai]

P2: Validate that messages is an array before spreading it. As written, a malformed or missing messages field will throw and return a 500 instead of a client error.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At apps/feature-request/app/api/chat/route.ts, line 43:

<comment>Validate that `messages` is an array before spreading it. As written, a malformed or missing `messages` field will throw and return a 500 instead of a client error.</comment>

<file context>
@@ -0,0 +1,78 @@
+          role: "system",
+          parts: [{ type: "text", text: SYSTEM_PROMPT }],
+        },
+        ...messages,
+      ],
+      model: {
</file context>