Skip to main content
Glama

OpenROAD MCP Server

A Model Context Protocol (MCP) server that provides tools for interacting with OpenROAD and ORFS.

About OpenROAD MCP

New here? Check out the Quick Start Guide to get your AI assistant analyzing designs in 5 minutes.

OpenROAD MCP eliminates the barrier between your AI assistant and physical design by connecting Claude, Cursor, and other MCP-compatible clients directly to the OpenROAD layout tools.

OpenROAD is the leading open-source, foundational application for semiconductor digital design, delivering an Autonomous, No-Human-In-Loop (NHIL) flow from RTL-GDSII. OpenROAD-flow-scripts (ORFS) is the fully autonomous flow built around it.

With this MCP server, your AI assistant can:

  • Execute Commands - Run interactive OpenROAD sessions with full PTY support.

  • Manage Sessions - Create, list, inspect, and terminate multiple physical design sessions.

  • Track History & Metrics - Access full command history and performance metrics for analysis.

  • Visualize Reports - List and read report images from ORFS runs directly in the chat.

Related MCP server: MCP Server

Demo

OpenROAD MCP Demo

Watch full demo video

Requirements & Installation

To use this MCP server, you need the server runtime, plus the underlying OpenROAD layout tools.

1. Server Runtime

  • Node.js 22+ is required to run the npx distribution.

2. OpenROAD

OpenROAD must be installed and available in your PATH.

3. OpenROAD-flow-scripts (ORFS)

ORFS is optional but highly recommended for complete RTL-to-GDS flows and report visualization.

Configuration

For platform-specific Node.js and C++ toolchain setup instructions, see the Cross-Platform Build Guide.

You do not need to clone this repo or pass path environment variables in the common case. The published npx package does not read a .env file.

On startup the server inherits the MCP client's environment, then fills PATH the same way which openroad would: current PATH, then your login-shell PATH, then common install locations (/opt/homebrew/bin, conda, local OpenROAD builds). ORFS_FLOW_PATH defaults to ~/OpenROAD-flow-scripts/flow, and is also detected when ORFS sits next to the openroad binary.

Supported MCP Clients

Here is the standard base configuration used across most clients:

{
  "command": "npx",
  "args": ["-y", "openroad-mcp"]
}

Find your specific client below for the exact configuration snippet and file location.

claude mcp add --transport stdio openroad-mcp -- npx -y openroad-mcp

Or add the standard config to .mcp.json / .claude/settings.json.

If a GUI-launched client still cannot find openroad, pass an override. Use command -v so you do not hard-code paths:

claude mcp add \
  --env PATH="$(dirname "$(command -v openroad)"):${PATH}" \
  --env ORFS_FLOW_PATH="${HOME}/OpenROAD-flow-scripts/flow" \
  --transport stdio openroad-mcp \
  -- npx -y openroad-mcp

Put --transport between --env and the server name so the CLI does not treat the name as another KEY=value pair.

Add the standard config to:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

  • Windows: %APPDATA%\Claude\claude_desktop_config.json

Add the standard config to .cursor/mcp.json.

Add to .vscode/mcp.json. Requires "type": "stdio":

{
  "servers": {
    "openroad-mcp": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "openroad-mcp"]
    }
  }
}

Add the standard config to ~/.codeium/windsurf/mcp_config.json.

Add the standard config to cline_mcp_settings.json (Cline) or .roo/mcp.json (Roo Code).

Add to your respective config.json under modelContextProtocolServers:

{
  "transport": {
    "type": "stdio",
    "command": "npx",
    "args": ["-y", "openroad-mcp"]
  }
}

Add to ~/.config/zed/settings.json:

{
  "context_servers": {
    "openroad-mcp": {
      "command": {
        "path": "npx",
        "args": ["-y", "openroad-mcp"]
      }
    }
  }
}

The server is available on the MCP Registry and via Docker:

docker run --rm -i ghcr.io/the-openroad-project/openroad-mcp:latest

Most other standard STDIO clients are fully supported. Refer to your tool's MCP setup guide.

Available Tools

Once configured, your AI assistant will have access to the following tools. For detailed parameters, schemas, and return formats, see the API Reference.

  • interactive_openroad_query

  • interactive_openroad_exec

  • create_interactive_session

  • list_interactive_sessions

  • terminate_interactive_session

  • inspect_interactive_session

  • get_session_history

  • get_session_metrics

  • list_report_images

  • read_report_image

Troubleshooting

  • The server fails to start: Ensure you have Node.js 22+. Older versions will fail.

  • Session creation fails: Confirm command -v openroad works in a terminal. The server inherits PATH and searches common install locations; if your prefix is unusual, pass PATH with --env as shown in the Claude Code section.

  • Commands rejected with CommandBlocked: You sent a state-modifying command to interactive_openroad_query. Use interactive_openroad_exec instead.

  • Report images not found: The server defaults to ~/OpenROAD-flow-scripts/flow. If ORFS lives elsewhere, set ORFS_FLOW_PATH in the MCP client's env block (not a .env file).

To get more detail, set LOG_LEVEL=DEBUG in the server's environment.

Development

Clone the repository. .env.example is a local-dev reference only; copy it to .env if you use direnv or similar. The server still reads process.env (the MCP client's env block), not the file.

Then run:

cd typescript
npm install
npm run build

Testing:

npm run test             # unit tests
npm run test:integration # integration tests
npm run test:performance # performance benchmarks

Linting & type checking:

npm run typecheck
npm run lint

Contributing

We welcome contributions! Please see CONTRIBUTING.md for detailed instructions on our development workflow and code standards.

License

BSD 3-Clause License. See LICENSE file.


Built with ❤️ by Precision Innovations

Available Tools

10 tools
create_interactive_sessionC

Create a new interactive OpenROAD session.

Inputs: session_id, command, env, cwd. Returns: result

get_session_historyC
Read-onlyIdempotent

Get command history for an interactive OpenROAD session.

Inputs: session_id (required), limit, search. Returns: result

get_session_metricsB
Read-onlyIdempotent

Get comprehensive metrics for all interactive OpenROAD sessions.

Inputs: no named parameters. Returns: result

inspect_interactive_sessionA
Read-onlyIdempotent

Get detailed inspection data for an interactive OpenROAD session.

Inputs: session_id (required). Returns: result

interactive_openroad_execA
Destructive

Execute a state-modifying OpenROAD command (set_*, create_*, read_*, write_*, flow commands).

Use this for loading designs, running placement/routing, applying constraints, and writing output files. Read-only commands are blocked — use interactive_openroad_query instead.

Inputs: command (required), session_id, timeout_ms. Returns: result

interactive_openroad_queryA
Read-only

Execute a read-only OpenROAD command (report_*, get_*, check_*, sta, help, etc.).

Use this for querying design state, generating reports, and inspecting timing. Commands that modify design state are blocked — use interactive_openroad_exec instead.

Inputs: command (required), session_id, timeout_ms. Returns: result

list_interactive_sessionsA
Read-onlyIdempotent

List all active interactive OpenROAD sessions.

Inputs: no named parameters. Returns: result

list_report_imagesB
Read-onlyIdempotent

List available report images from ORFS runs organized by stage.

Inputs: platform (required), design (required), run_slug (required), stage. Returns: result

read_report_imageC
Read-onlyIdempotent

Read a report image and return base64-encoded data with metadata.

Inputs: platform (required), design (required), run_slug (required), image_name (required). Returns: result

terminate_interactive_sessionA
DestructiveIdempotent

Terminate an interactive OpenROAD session.

Inputs: session_id (required), force. Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.5/5.0

    Scored across 10 tools

    Disambiguation4/5

    The query/exec split is clearly disambiguated by read-only vs state-modifying behavior with explicit cross-references. Session management, command execution, and report image tools each have distinct targets, though inspect_interactive_session and get_session_metrics have slight semantic overlap.

    Naming Consistency4/5

    Most tools follow a clear verb_noun pattern: list/create/terminate/inspect/get/read. However, interactive_openroad_query and interactive_openroad_exec put the verb at the end and do not match the otherwise consistent session/report naming pattern.

    Tool Count5/5

    Ten tools is well-scoped for the server's purpose: command execution, session lifecycle, session introspection, and report image access. Each tool has a clear role and no redundant additions inflate the count.

    Completeness4/5

    The tool surface covers session creation/listing/termination, read/write command execution, history/metrics, and report image retrieval. Minor gaps exist, such as no explicit session attach or live output streaming, but the core workflows are fully usable without dead ends.

    Maintenance

    ActivityActive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      MCP Server simplifies the implementation of the Model Context Protocol by providing a user-friendly API to create custom tools and manage server workflows efficiently.
      39 npm
      4
      MIT
    • F
      license
      Not graded
      quality
      D
      maintenance
      A server implementation of the Model Context Protocol (MCP) that provides REST API endpoints for managing and interacting with MCP resources.
      -