# MCP (Model Context Protocol)

MCP: The open protocol that lets a chat assistant like Claude or ChatGPT read and act on your CodeHerder workspace directly, without a custom integration.

Source: https://codeherder.com/glossary/mcp/

Glossary · The machinery

The open protocol that lets a chat assistant like Claude or ChatGPT read and act on your CodeHerder workspace directly, without a custom integration.

## What it is

MCP, short for Model Context Protocol, is a shared standard for letting an AI assistant call out to an external tool or service. Instead of every product building its own one-off plugin system, a service exposes an MCP connector, and any MCP-compatible assistant can use it the same way. It's plumbing more than a product in its own right, closer to a common plug shape than to an app.

## Why it matters to you

If you already spend time in claude.ai or ChatGPT, MCP is what lets you check on a task or file a new one from inside that same conversation, without opening a separate app. For a non-technical founder, it means the AI chat window you already use can double as a window into your CodeHerder workspace, so filing a piece of work is as easy as describing it to an assistant you already talk to daily.

## How it works in CodeHerder

CodeHerder ships an MCP connector that a claude.ai or ChatGPT conversation can use to read and write on your behalf, scoped to what that connection is allowed to touch. Exactly what it can do, what it can't reach, and what to do if something looks wrong, is covered in the MCP tools doc, along with how to set the connection up the first time.

- [MCP tools →](https://codeherder.com/docs/mcp-tools/)

## What people get wrong

MCP can sound like it hands the assistant control over what work gets done. It's the opposite: MCP just gives the assistant a scoped set of actions it's allowed to take on your behalf. The same tasks and workflow stages still apply, and so do the approval gates. Filing a task by chatting with an assistant doesn't skip review any more than filing it through the web app would. It's simply a different front door onto the same platform, and the same person still has to approve the risky stages either way.

Keep going

## Related terms

[AI coding agent Software that reads a task written in plain language and produces working code changes on its own.](https://codeherder.com/glossary/ai-coding-agent/) [Agent orchestration Running more than one coding agent against a shared, tracked backlog, instead of leaving it to whoever opened a terminal.](https://codeherder.com/glossary/agent-orchestration/)

[← See every term](https://codeherder.com/glossary)