Vivado MCP Native

English | 简体中文

A cross-platform Model Context Protocol (MCP) server for AMD/Xilinx Vivado. It lets MCP-compatible AI clients start and manage Vivado, open projects, run synthesis and implementation, inspect timing and utilization, control simulation, and execute Tcl commands.

This project uses a persistent subprocess Tcl session that works natively on Windows and Linux. It does not depend on the Unix-only pexpect transport.

Highlights

• Native launch of Windows vivado.bat and vivado.cmd files.
• Direct Linux executable launch without an intermediate shell.
• One persistent Vivado Tcl process reused across MCP calls.
• UUID-framed command protocol that separates stdout, Tcl return values, return codes, and error stacks.
• UTF-8 hex transport for quotes, braces, backslashes, multiline Tcl, Unicode, and Windows paths.
• Full process-tree cleanup after command timeouts.
• Vivado discovery through VIVADO_PATH, PATH, and common installation directories.
• A vivado-mcp-native-doctor command for checking the local Python-to-Vivado connection.

Requirements

Installation

From PyPI — recommended

Install the isolated command-line application with pipx:

py -m pip install --user --upgrade pipx
py -m pipx ensurepath
py -m pipx install vivado-mcp-native

Linux:

python3 -m pip install --user --upgrade pipx
python3 -m pipx ensurepath
python3 -m pipx install vivado-mcp-native

Installed commands:

• vivado-mcp-native — start the MCP server;
• vivado-mcp-native-doctor — check the local Vivado connection;
• vivado-mcp-win and vivado-mcp-win-doctor — compatibility aliases.

Install the latest source from GitHub

py -m pipx install --force "https://github.com/Arthurzxy/vivado_mcp_native/archive/refs/heads/master.zip"

Dedicated virtual environment

$venv = "$env:LOCALAPPDATA\vivado-mcp-native"
py -3.11 -m venv $venv
& "$venv\Scripts\python.exe" -m pip install --upgrade pip
& "$venv\Scripts\python.exe" -m pip install --upgrade vivado-mcp-native

Official MCP Registry

The server identity used by the official MCP Registry is:

io.github.Arthurzxy/vivado-mcp-native

Registry metadata is stored in server.json. The PyPI package uses the hidden mcp-name marker near the top of this README for package ownership verification. Version 0.2.1 is the first release prepared for official Registry publication.

Select the Vivado installation

VIVADO_PATH and the start_session tool's vivado_path argument may point to:

• the complete vivado.bat, vivado.cmd, vivado.exe, or Linux vivado launcher;
• the Vivado bin directory;
• the Vivado version directory.

Example:

$env:VIVADO_PATH = "D:\Xilinx\Vivado\2025.2\bin\vivado.bat"

Without an explicit path, the server checks PATH and common Windows locations such as:

C:\Xilinx\Vivado\*\bin\vivado.bat
C:\AMD\Vivado\*\bin\vivado.bat
C:\Program Files\AMD\Vivado\*\bin\vivado.bat

Check the Vivado connection

vivado-mcp-native-doctor --vivado-path "D:\Xilinx\Vivado\2025.2\bin\vivado.bat"

The path may also be a Vivado bin directory or version directory. Machine-readable output is available with --json:

vivado-mcp-native-doctor --vivado-path "D:\Xilinx\Vivado\2025.2" --json

The diagnostic command resolves the launcher, starts a persistent Tcl session, queries Vivado and Tcl versions, checks Tcl command transport and Unicode handling, verifies session health, and closes Vivado cleanly. It does not open or modify a user project.

Configure an MCP client

Find the installed executable:

(Get-Command vivado-mcp-native).Source

Example configuration for a pipx installation:

{
  "mcpServers": {
    "vivado": {
      "command": "C:\\Users\\you\\.local\\bin\\vivado-mcp-native.exe",
      "env": {
        "VIVADO_PATH": "D:\\Xilinx\\Vivado\\2025.2\\bin\\vivado.bat"
      }
    }
  }
}

Example configuration for a virtual environment:

{
  "mcpServers": {
    "vivado": {
      "command": "C:\\Users\\you\\AppData\\Local\\vivado-mcp-native\\Scripts\\python.exe",
      "args": ["-m", "vivado_mcp"],
      "env": {
        "VIVADO_PATH": "D:\\Xilinx\\Vivado\\2025.2\\bin\\vivado.bat"
      }
    }
  }
}

Using an absolute executable or Python path avoids depending on the MCP client's PATH environment.

Capabilities

Architecture

MCP-compatible client
        │ MCP / stdio
        ▼
Vivado MCP server
        │ persistent subprocess Tcl session
        ▼
AMD/Xilinx Vivado -mode tcl

Each command is transported as UTF-8 hex and wrapped with unique markers. The server extracts stdout, the Tcl result, return code, and error stack without relying on the localized Vivado% prompt.

Troubleshooting

Vivado does not start

Verify the launcher directly:

& "D:\Xilinx\Vivado\2025.2\bin\vivado.bat" -mode tcl

Then run vivado-mcp-native-doctor and check the path, installation, permissions, environment, and license configuration.

Chinese output is corrupted

Override the Windows output decoder when necessary:

$env:VIVADO_MCP_OUTPUT_ENCODING = "gbk"

Use utf-8, gbk, or the encoding that matches the Vivado Tcl console.

A command times out

The server terminates the Vivado process tree after a timeout to avoid a desynchronized session. Start a new session and use a larger timeout for long synthesis or implementation runs.

More Windows guidance is available in WINDOWS_INSTALL.md.

Security

run_tcl can execute arbitrary Tcl with the current user's permissions, including file operations and external process launches. Expose this server only to trusted MCP clients and review high-impact commands and paths.

License

This project is available under the MIT License.

Acknowledgments

• Original project created by Corey Hahn.
• Built on the Model Context Protocol.
• Integrates with AMD/Xilinx Vivado.