Start for free Book a free session

MCP server

Sigasi ships an MCP server that exposes its semantic engine to AI coding agents through the Model Context Protocol  (MCP). Instead of guessing, your agent can deterministically format HDL, read Sigasi’s real diagnostics, and explore your project’s structure and use that feedback to iterate on and fix the code it generates.

This turns Sigasi into the deterministic checker in an agentic loop: the agent writes or edits Verilog, SystemVerilog, or VHDL, formats it, asks Sigasi what’s wrong, and fixes the reported problems—repeating until the file is clean.

Requirements

  • Requires a configured Sigasi project. The tools operate on files within your project, so diagnostics, context, and symbol search only return meaningful results once your project is loaded.
  • The MCP server runs inside the Sigasi extension. A VS Code window with your project open has to be running. For agents running without an editor, such as in CI, use the Sigasi CLI instead.
  • Works out of the box with GitHub Copilot in VS Code. Other agents (Claude Code, Codex, etc.) connect after a one-time registration.
  • The sigasi.enableMcpServer setting is enabled (it is by default).

Registration

GitHub Copilot

The MCP server is built into the Sigasi extension and is registered automatically with VS Code’s built-in MCP registry. There is nothing to install or register separately: once Sigasi is installed and your project is loaded, the Sigasi tools become available to your AI agent via Copilot. In the agent’s tool picker they appear under Sigasi Visual HDL.

The Sigasi Visual HDL MCP tools listed in the agent's tool picker

Other agents

Agents that do not read VS Code’s MCP registry are registered once, from the MCP registration page. Open the Command Palette ( Ctrl+Shift+P ) and run Sigasi: Open MCP Registration Page, or follow the MCP Registration link in the Sigasi settings.

The Sigasi MCP registration page showing registration snippets per AI agent

The page lists the agents that Sigasi supports (Claude Code, Codex, etc.). For each of them, it shows the exact command to register Sigasi, already filled in with the location of the Sigasi MCP server on your machine. Copy it with , or let Sigasi write the entry into the agent’s configuration file for you with .

Registration is permanent, but the connection is not: because the server lives inside the extension, a registered agent only reaches Sigasi while your project is open.

The Sigasi MCP tools return their results in the structuredContent field of the tool response. AI agents that don’t read structuredContent, such as Kiro CLI and Devin, may not work correctly with Sigasi’s MCP server, even though they can register it as an MCP server.

Getting the agent to use the tools

Keep in mind that having the tools available does not guarantee the agent uses them: the AI decides for each request which tools, if any, to invoke. You can steer this decision by mentioning Sigasi explicitly in your prompt, for example “Fix the problems in fifo.vhd using Sigasi”. To make this the default behavior, add an instruction to your project’s AGENTS.md (or your agent’s custom instructions file), such as “Always use the Sigasi tools to check, fix, and format HDL files.”

Available tools

The server provides the following tools to the agent:

CapabilityWhat the agent can do
Get file contextGet project information for a file: its project, targets, libraries, and dependencies.
Get diagnosticsRetrieve Sigasi’s diagnostics for a specific file.
Get project diagnosticsRetrieve diagnostics for the whole project that contains a file.
Search project symbolsSearch design units and symbols across the Sigasi project.
Format documentFormat a Verilog, SystemVerilog, or VHDL file in place.

Example: fixing a file with an agent

Ask your agent to clean up a file and let it drive the tools. For a prompt like “Fix the errors in fifo.vhd”, the agent typically:

  1. Calls Get diagnostics on the file to see what Sigasi reports.
  2. Uses Search project symbols and Get file context to understand the surrounding design—its libraries, dependencies, and where symbols are defined.
  3. Edits the file to resolve the reported problems.
  4. Calls Format document to apply Sigasi’s formatting.
  5. Calls Get diagnostics (or Get project diagnostics) again to confirm the problems are gone, repeating the loop if needed.

Because Sigasi’s analysis runs at interactive speed, the agent gets this feedback immediately, keeping the loop tight.

An AI agent chat exercising the Sigasi diagnostics and formatting tools