What is LabVIEW MCP?
LabVIEW MCP, published by the engineering consultancy Zühlke, lets an MCP client work with LabVIEW code on a Windows machine. It returns a VI's block diagram as text, generates new VIs, edits existing ones in place, runs a VI as a top-level VI and reads its outputs back, executes build specifications, and searches installed palettes and NI's shipping examples. It ships as one Windows executable, LabVIEWMCP.exe, with 121 tools, and as a Claude Code plugin that adds eight LabVIEW agents for tasks such as VI generation, editing, class and DQMH module creation, documentation and unit tests.
It talks to LabVIEW through two engines: NI's open-source grpc-labview server, which LabVIEW's AI add-on runs on 127.0.0.1, and a bundled copy of pylabview that reads and rewrites the .vi file directly. Of several projects named labview-mcp on GitHub it is the most starred, at about 154, MIT-licensed, and released v3.0.0 on 17 November 2026. It is not affiliated with NI or Emerson.
How to install LabVIEW MCP
claude plugin marketplace add Zuehlke/labview-mcp && claude plugin install labview-mcp@zuehlke-labviewIs LabVIEW MCP worth it?
LabVIEW MCP solves a real gap: a .vi is a binary file that a language model cannot read or write, so assistants have mostly skipped LabVIEW. This server turns a VI into text an agent can read and produces real .vi files from text, using LabVIEW's own AIXML generator where possible and pylabview where the generator cannot reach. The authors measured that only 15% of 900 VIs in a production codebase can be regenerated through AIXML, mainly because 70% call the project's own subVIs, so editing existing code usually goes through pylabview.
The documentation is unusually frank about risk. The README labels the project research-grade, measured by its authors on one station and never near a production validation cycle. The underlying lvai.LVAI interface is not a published NI API, so a LabVIEW upgrade can break it. According to the safety guide, the write tools overwrite a .vi without asking, regenerating a VI discards its layout, decorations and icon, one class of edit made LabVIEW.exe terminate on load and was removed, and the ApplyAIXMLToVI call reports success while writing nothing. The local gRPC port has no TLS and no authentication, so anything on the machine that can reach it can drive LabVIEW.
The catch for test and control engineers is that running a VI is running code against whatever that VI controls. The project warns in its own words of a generated VI that drives real hardware into a wall. Read tools carry readOnlyHint and can be allow-listed, write tools carry destructiveHint and prompt every time, and the Claude Code plugin ships an allow-list that keeps those prompts on. That is the only gate; nothing in the server knows whether a VI talks to a DAQ card, a motion controller or a simulation, so every generated VI needs review and sign-off by an engineer who owns the rig.
Setup is easy by the standard of engineering MCP servers: two Claude Code commands install a prebuilt binary, and other clients are covered in the install guide. Our take: as a reader and documenter of an inherited LabVIEW codebase it earns its place today; as a code generator, treat it as a junior who works only on a branch.
Where it earns its place
A test engineer inheriting a 900-VI production test system
Understand what an undocumented VI hierarchy does before changing it
Have the agent read VIs and the .lvproj as text and summarise each subVI, using only allow-listed read tools. Nothing is written, so the risk is a wrong explanation, not a broken project.
A LabVIEW team lead at a small automation integrator
Produce a Word document for a class library before a customer handover
Run the bundled documentation agent over the library to get a structure diagram and a section per public VI, then edit the draft. It saves hours of manual write-up; the text still needs checking against the code.
A lab engineer prototyping a data-logging VI
Draft a VI that reads a CSV file and sorts it, then check it runs
The agent looks up NI examples, writes AIXML, generates the VI and runs it to read outputs back, prompting before each write. Work on a branch, and do not run generated VIs connected to live instruments until an engineer has reviewed them.
Before you commit
- Setup
- Windows x64, LabVIEW 2020 or newer and an MCP client. In Claude Code 2.1.224 or later, two plugin commands install the prebuilt LabVIEWMCP.exe, the eight agents and an allow-list; other clients use the install guide. On LabVIEW 2026 Q3 it can use NI's gRPC service, on older versions pylabview and LabVIEW's ActiveX server.
- Integrations
- 121 tools over stdio for Claude Code, Claude Desktop, Cursor, Codex, Copilot or a local model. Talks to LabVIEW through NI's grpc-labview on 127.0.0.1 and to .vi files through bundled pylabview; searches installed palettes and NI examples, including OpenG, MGI and JKI where present.
- Data and privacy
- All four processes run locally and nothing goes to the internet except what your MCP client sends to its model provider, which includes VI contents. The gRPC port on 127.0.0.1 has no TLS or authentication. Write tools overwrite files without backups; the README says version control is the only undo.
- Cost at scale
- Free under MIT. Each seat needs whatever LabVIEW licence it already uses; NI's pricing is not covered in the project's documentation.
- Limits
- Only 15% of VIs in the authors' 900-VI sample can be regenerated through AIXML; ApplyAIXMLToVI reports success but writes nothing; LabVIEW upgrades may break the private interface. Tested by its authors on one station.
What’s good
- 121 tools covering read, generate, edit, run and build
- Eight bundled LabVIEW agents in the Claude Code plugin
- Read and write tools annotated; writes always prompt
- pylabview engine edits VIs without a LabVIEW licence
What’s not
- Relies on an unpublished NI interface that upgrades can break
- Write tools overwrite VIs and drop layout and icons
- Local gRPC port has no TLS or authentication
- Windows x64 only; research-grade by its own account
Best for
LabVIEW developers on Windows who want an agent to explain, document and draft VIs in a version-controlled project, with a person approving every write.
LabVIEW MCP alternatives
LabVIEW MCP: common questions
- Is LabVIEW MCP worth it?
- A serious, candid attempt to give agents a way into LabVIEW, fit for supervised work on copies under version control and nowhere near a production test stand unsupervised. We rate it 3.8 out of 5.
- How much does LabVIEW MCP cost?
- Free and open source (MIT). Needs LabVIEW 2020 or newer, which is commercial NI software licensed separately; the pylabview engine works without a LabVIEW licence.
- What are the downsides of LabVIEW MCP?
- Relies on an unpublished NI interface that upgrades can break; write tools overwrite VIs and drop layout and icons; local gRPC port has no TLS or authentication; windows x64 only; research-grade by its own account.
- Who is LabVIEW MCP best for?
- LabVIEW developers on Windows who want an agent to explain, document and draft VIs in a version-controlled project, with a person approving every write.
- How much does LabVIEW MCP cost for a team or in production?
- Free under MIT. Each seat needs whatever LabVIEW licence it already uses; NI's pricing is not covered in the project's documentation.
- Is LabVIEW MCP safe for company data?
- All four processes run locally and nothing goes to the internet except what your MCP client sends to its model provider, which includes VI contents. The gRPC port on 127.0.0.1 has no TLS or authentication. Write tools overwrite files without backups; the README says version control is the only undo.
- What are the best LabVIEW MCP alternatives?
- The closest alternatives we have reviewed are LabMCP and Arduino MCP Server.
Answers checked Nov 18, 2026.