Features & Tools

MCP Overview

Model Context Protocol — extend DevcoreAI with custom tools and external services.

What is MCP?

Model Context Protocol (MCP) is an open standard that allows DevcoreAI to connect to external tools, databases, and APIs through a standardized interface. Think of MCP servers as plugins that give the agent new superpowers.

Each MCP server exposes a set of tools — actions the agent can invoke. For example, a Slack MCP server might expose tools like send_message, list_channels, and search_messages.

Data sources

Databases, file systems, data warehouses, REST APIs

External services

Slack, GitHub, Figma, Notion, Linear, Jira

Custom tools

Internal APIs, build systems, infrastructure tools

How MCP Servers Connect

DevcoreAI supports three MCP transport types. Most community servers use stdio.

stdio

Subprocess — MCP server runs as a child process. Most common. Works offline.

Recommended

HTTP (SSE)

Remote server over HTTP with Server-Sent Events. Good for cloud services.

WebSocket

Real-time bidirectional. Good for interactive/stateful tools.

Quick Setup

Install a popular server directly from the MCP panel in DevcoreAI, or configure one manually via mcp_config.json.

One-click install (in DevcoreAI IDE)

1. Open DevcoreAI → click the MCP Servers button (server icon)

2. Browse the plugin store

3. Click Configure next to any server

4. Follow the OAuth or API key prompts

✓ Server is now available as tools in all agent tasks

mcp_config.json — manual config example

{
  "mcpServers": {
    "postgres": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres"],
      "env": {
        "DATABASE_URL": "postgresql://user:pass@localhost/mydb"
      }
    }
  }
}

Full MCP setup guide

Popular MCP Servers

Figma

5 tools

Inspect design specs, read component properties, and get design tokens directly in the agent context.

Slack

7 tools

Read channels, send messages, and search workspace conversations from within DevcoreAI tasks.

GitHub

9 tools

List PRs, read issues, create branches, and trigger actions without leaving your editor.

PostgreSQL

6 tools

Query your database, inspect schemas, and run migrations with full context in the agent.

Playwright

4 tools

Automate browser testing — click, navigate, screenshot, and validate web app behavior.

Sequential Thinking

3 tools

Structured reasoning chains for complex multi-step problem-solving tasks.

Security & Approvals

Every MCP tool call is shown to you before execution. You can approve or reject each invocation individually, or configure trusted servers for auto-approval.

Each tool call shows the exact parameters before execution
Sensitive values (API keys, passwords) are masked in the UI
Auto-approval can be scoped to specific servers or tools
All MCP calls are included in the task checkpoint history

Building Custom MCP Servers

MCP servers can be written in any language with a transport adapter. The SDK is available for TypeScript and Python. Wrap any internal API, build tool, or custom service as an MCP server and the agent will be able to use it naturally.

TypeScript SDK Python SDK

← Checkpoints Workflow Hooks