MCP server

Connect an AI assistant to EngineHall. The hosted server reads your account and saved projects on enginehall.com and works in your development projects. The local servers review design files on your own computer.

Documentation

Hosted server

The hosted server lets an assistant read what you can see on enginehall.com: your account and plan, your workspaces, and the saved projects of paid workspaces with their reports and findings. It also answers questions about the rules and the finding format. In development projects it reads open points, decisions, meetings and documentation, and drafts and updates them within the limits below. From a repository connected to a development project it reads text files such as firmware sources. It never receives design files and never returns binary ones.

Server URL

https://enginehall.com/mcp

Connected to https://pcbastudio.com/mcp before the rename? That address keeps working until 30 April 2027. Reconnect with https://enginehall.com/mcp before then.

To review a design folder, use a local server: Local servers

Connect your assistant

Every assistant uses the same URL. It registers itself, sends you to enginehall.com to sign in, and asks you to approve access.

Claude (web and desktop): Settings, Connectors, Add custom connector. Paste the URL, then Connect. On Team and Enterprise plans an owner adds the connector once under Organization settings, Connectors, and each member connects their own account.
Claude Code: Add the server, then run /mcp in Claude Code and choose Authenticate.
claude mcp add --transport http enginehall https://enginehall.com/mcp
ChatGPT: Settings, Apps, Advanced settings. Turn on Developer mode, then create an app with the URL and OAuth authentication.
Codex: Add the server, then sign in:
codex mcp add enginehall --url https://enginehall.com/mcp codex mcp login enginehall
Cursor: Add the server to your Cursor MCP configuration:
~/.cursor/mcp.json
{ "mcpServers": { "enginehall": { "url": "https://enginehall.com/mcp" } } }
VS Code: Add the server to the MCP configuration of your workspace:
.vscode/mcp.json
{ "servers": { "enginehall": { "type": "http", "url": "https://enginehall.com/mcp" } } }

Tools

Read tools only read. Draft tools create items that only your organisation sees, and change tools edit open points assigned to you or created by you, and internal pages. Every draft and change tool first returns a preview and changes nothing. Only when the assistant calls it again with confirm: true, after you approved the preview, is the change made.

Sharing an item with the whole project, changing its visibility, agreeing a decision, confirming a meeting, approving and inviting people have no tool. The assistant gives you a link, and you do it yourself in the browser.

Tool	What it does	Access	Needs confirm
get_account	Your email, display name, plans, and how many unfinished saves you hold.	Read	No
list_workspaces	Your workspaces with plan, your role, member count, saved projects, and what a save may hold.	Read	No
list_saved_projects	Saved projects in your workspaces, newest first, with finding counts. Filters by workspace, name and status.	Read	No
get_saved_project	One saved project: author, options, finding summary, coverage, and the names and sizes of the files it was reviewed from.	Read	No
list_findings	Findings of a saved project in compact form, filtered by severity, rule, domain or file.	Read	No
get_finding	One finding in full: location, quoted evidence, consequence and recommendation.	Read	No
get_report	The stored report as JSON, SARIF, Markdown or text.	Read	No
list_rules	The review rules with severity, detector and the inputs each needs.	Read	No
get_rule	One rule, and why it can appear in coverage.skipped.	Read	No
get_schema	The JSON Schema of a finding or of a baseline file.	Read	No
get_local_mcp_setup	How to install and connect the local servers.	Read	No
list_projects	Your development projects with key, your role and organisation.	Read	No
project_overview	One development project: organisations, open points by kind and status, overdue items, release blockers, decisions and the latest meetings.	Read	No
list_open_points	Actions, questions and risks with their ids (KEY-A-12), filtered by kind, status, owner and visibility.	Read	No
get_open_point	One open point by id: text, owner, due date, evidence, source meeting and comments.	Read	No
list_decisions	Decisions and ADRs with status, agreement date and text.	Read	No
list_meetings	Meetings, newest first, with attendees and summary.	Read	No
get_page	A documentation page as Markdown, or the list of pages.	Read	No
search	Search the open points, decisions, meetings and pages of a project.	Read	No
list_repositories	The repositories connected to a project, with provider, URL and default branch.	Read	No
get_file	One text file from a connected repository at a branch or commit, size-capped.	Read	No
add_open_point	Create an action, question or risk. It is always internal to your organisation.	Draft	Yes
comment	Comment on an open point, decision, meeting or page. The preview says who will see it.	Draft	Yes
draft_page_change	Add a revision to an internal page or create a new internal page. For a shared page it only returns a proposal.	Draft	Yes
draft_meeting_summary	Create an internal, unconfirmed meeting with its summary.	Draft	Yes
propose_decision	Create an internal decision or ADR with the status proposed.	Draft	Yes
update_open_point	Change the title, text, owner, due date or status of an open point assigned to you or created by you.	Change	Yes
close_open_point	Close an open point assigned to you or created by you, with a closing note or evidence.	Change	Yes
apply_page_change	Replace the text of an internal page, refused if the page changed since you read it.	Change	Yes
list_releases	List the releases of a project with version, status, checklist progress and release date.	Read	No
get_release	A release with notes, commits, check run, checklist, approvals per organisation and what still blocks it.	Read	No
draft_release	Create an internal draft release on the latest usable check run and its commit, with the project's checklist.	Draft	Yes
list_check_runs	Check runs of a development project, newest first: commit, source, finding counts and coverage gaps.	Read	No
get_check_run	One check run: coverage and its findings with rule, severity, location and evidence.	Read	No
list_reviews	Design reviews with human id, status, subject, reviewer verdicts and unresolved threads.	Read	No
get_review	One design review: subject, reviewers and verdicts, comment threads with their anchors.	Read	No
comment_on_review	Start an anchored thread on a design review or reply in one. It gives no verdict; reviewers approve on the page.	Draft	Yes

Safety model

The app acts as you. Every query runs with your own permissions on enginehall.com, so it sees exactly the workspaces and saved projects you can open in the browser, and nothing of other customers.
The tools for saved projects only read. Saving, rerunning and deleting a project, inviting members and billing stay in the browser.
In development projects, what an assistant creates stays internal to your organisation, and nothing changes before you approve its preview. Your project role still applies: a viewer cannot write, and a reviewer only updates open points assigned to them.
Design files never pass through the hosted server. A saved project is described by the names and sizes of its files, never their contents. A design folder is reviewed by a local server. get_file returns text files of a connected repository only, read with the repository's stored token after your own access is checked; binary files such as schematics are refused.
The token the app receives carries your account's permissions on enginehall.com, not only the MCP tools. Approve only apps you trust and revoke the ones you no longer use.
Each tool call is logged with your account id, the app's client id and the tool name. Arguments and results are not logged.

Revoke access

Your account page lists every app you approved. Revoking one ends its sessions and refresh tokens at once, and the app has to ask you again to reconnect. Connected apps

Limits

About 120 tool calls per minute for each account and app. A call over the limit returns an error that says how long to wait.
Lists return at most 100 entries per call. Pass offset for the next page.
get_report returns renderings up to 1,000,000 characters. Read larger reports with list_findings and get_finding.
get_file returns text files up to 200,000 characters, from files of at most 5 MB.
Access tokens are valid for one hour, and the app refreshes them while access stays approved.

Technical details

Transport: Streamable HTTP, stateless. Each request gets one JSON response. There is no event stream, so a GET answers 405.
Authorization: OAuth 2.1 with PKCE and dynamic client registration. Discovery: https://enginehall.com/.well-known/oauth-protected-resource/mcp.
Protocol versions: 2025-11-25, 2025-06-18 and 2025-03-26.

Local servers

Schematics, CubeMX files and firmware are reviewed on your computer, and the command line and the VS Code extension send no design content anywhere. Their MCP servers are programs your assistant starts on your computer. Install enginehall first and activate its virtual environment, or give the client the full path to the enginehall program. Install

enginehall mcpmacOS, Linux, Windows: Reviews a project folder: finds the schematic, the .ioc and the firmware, runs every rule, and returns findings with evidence and coverage. Tools: pcba_project_inspect, pcba_review_run, pcba_review_get_finding, pcba_altium_pinmap, pcba_cubemx_inspect, pcba_cubemx_validate.
.mcp.json
{ "mcpServers": { "enginehall": { "command": "enginehall", "args": ["mcp"] } } }
enginehall altium mcpWindows, with Altium Designer running: Creates and edits Altium projects with typed operations. Every write is planned and previewed, applied with the plan's hash after you approve it, and verified against the netlist.
.mcp.json
{ "mcpServers": { "enginehall-altium": { "command": "enginehall", "args": ["altium", "mcp"] } } }
enginehall-extensionVS Code and Cursor: Comes with the VS Code extension. Reads one .SchDoc sheet (schdoc_* tools) and maintains the Markdown project tracker (tracker_* tools). Register it with the command "EngineHall: Set up MCP server for AI chat".

In Claude Code, add the review server like this:

Claude Code

claude mcp add enginehall -- enginehall mcp

Agent skills teach an assistant which enginehall command to run and how to read the result. Download them on these pages: Altium, Fusion

Support

For questions or a connection that does not work, write to us. Never send an access token or design files. software@newmatik.com