MCP server for website feedback explained.

What is the Model Context Protocol (MCP)?

According to the official Model Context Protocol documentation, MCP is an open-source standard created by Anthropic that connects AI applications to external data sources and tools. MCP acts like a universal port for AI assistants, providing a standardized JSON-RPC 2.0 communication layer.

MCP defines three core primitives:

• Tools: Executable functions that allow AI models to perform actions and fetch dynamic data.
• Resources: File-like data streams that provide contextual information to models.
• Prompts: Reusable prompt templates that guide model interactions.

AI hosts such as Claude Code, Cursor, VS Code, and Codex connect to MCP servers either locally through standard input/output (stdio) or remotely using Streamable HTTP (HTTP POST with SSE).

How coding agents resolve website feedback over MCP

When human reviewers leave feedback on a website, an AI coding agent can resolve comments through an automated feedback loop:

1. Fetch open feedback: The agent calls an MCP tool to list unresolved threads for a specific project, route, or branch.
2. Inspect element context: The agent retrieves the thread details, including the page URL, CSS selector, visible text, and optional console logs.
3. Locate code in repository: Using its local code search tools, the agent finds the component, template, or style file corresponding to the selector and text.
4. Implement and test: The agent makes the required code edit, runs local tests, and verifies the layout.
5. Reply and resolve: The agent posts a reply explaining the change and invokes an MCP tool to mark the thread resolved.

Exact tools in the pinthread MCP server

The pinthread MCP server exposes dedicated tools for coding agents:

• list_threads: List feedback threads for a project with filters for status, branch, route, or assignee.
• get_thread: Fetch a full thread with comments, element selector, visible text, and attachment metadata.
• get_page_context: Retrieve all open threads and element anchors for a specific URL route.
• reply_thread: Post a comment in a thread explaining code adjustments or asking clarifying questions.
• resolve_thread: Mark a thread resolved after verifying the fix in code.
• ask_thread: Post a question to human reviewers and mark the thread as waiting for input.
• dismiss_thread: Close feedback with an explanation when no code modification is needed.
• list_events: Poll new comments and replies using a cursor to trigger agent workflows.
• mark_suggestion_applied: Apply suggested text changes, add a note, and close the item.
• whoami: Inspect current project key, token name, and permissions.

Connecting coding agents: setup and pointers

Setting up an MCP feedback server takes two quick steps:

1. Create a project token in your dashboard under Agent access with read or write scope.
2. Add the remote server URL to your agent configuration. For Claude Code, run claude mcp add --transport sse pinthread https://pinthread.dev/mcp. For Cursor, add the server to .cursor/mcp.json.

See our MCP server documentation, read the agent prompt guide, or follow the Claude Code walkthrough for full configuration details.

How pinthread helps with MCP feedback

pinthread hosts a remote MCP server at https://pinthread.dev/mcp. AI coding agents like Claude Code, Cursor, and Codex connect over the web to read real reviewer comments on staging and production builds without needing local browser extensions or manual copy-pasting.

Check the installation guide to add the review widget to your site.

Questions