• Site: https://mcp.lightbringer.com/mcp

1. Setup

Before you start: MCP access must be enabled for your organisation in the Lightbringer app. Go to Advanced Settings, turn on "Enable MCP access for this organisation," and click "Save changes." Without this, the connector's sign-in will fail.

1.1 Claude (web and desktop)

Once it's in the Claude directory (easiest): when the Lightbringer listing is approved in the Claude Connectors Directory, you'll be able to enable it in one click: Settings → Connectors → browse the directory → Lightbringer → Connect, then sign in. No URL needed.

Add it now as a custom connector:

1. In Claude, open Settings → Connectors (on claude.ai: Customize → Connectors).
2. Click Add custom connector.
3. Enter:
   • Name: Lightbringer
   • URL: https://mcp.lightbringer.com/mcp
4. Click Connect. Claude opens Lightbringer's sign-in. If sign-in fails, check that MCP access is enabled for your organisation (see Before you start).
5. Sign in with email, Google, or Microsoft — whichever your Lightbringer account uses.
6. On the consent screen, approve the requested access (read + write) and select the organisation you want to work in. Access is scoped to that single organisation.
7. You're connected. Ask Claude "Who am I on Lightbringer?" to confirm — it should return your user, organisation, and permissions.

1.2 Claude Code

Claude Code is Anthropic's agent for the terminal and IDEs. One command adds Lightbringer for all your projects:

claude mcp add --transport http --scope user lightbringer https://mcp.lightbringer.com/mcp

Then, inside a Claude Code session:

1. Run /mcp and select lightbringer.
2. Choose Authenticate. Your browser opens Lightbringer's sign-in — sign in with email, Google, or Microsoft, approve the requested access, and select your organisation.
3. Back in the terminal, ask "Who am I on Lightbringer?" to confirm.

If your session ever loses access, run /mcp again and choose Re-authenticate. Omit --scope user if you'd rather enable Lightbringer for the current project only.

1.3 ChatGPT

Custom MCP connectors with full read + write access require ChatGPT's Developer mode (beta), available on paid plans. Turn it on under Settings → Connectors → Advanced → Developer mode. On Business, Enterprise, and Edu plans, a workspace admin must first allow custom connectors under workspace permissions.

1. In ChatGPT, go to Settings → Connectors and click Create.
2. Enter:
   • Name: Lightbringer
   • Description: Patent and invention management — search, draft, review, and submit invention disclosures.
   • MCP server URL: https://mcp.lightbringer.com/mcp
   • Authentication: OAuth
3. Click Create and complete the sign-in: sign in with email, Google, or Microsoft, approve the requested access, and select your organisation.
4. In a new chat, click the "+" in the message box, enable the Lightbringer connector under Developer mode, and ask "Who am I on Lightbringer?" to confirm.

Note: without Developer mode, ChatGPT limits connectors to search-and-fetch research. You'll be able to search and read your Lightbringer portfolio (including in deep research), but not draft, create, or submit disclosures.

1.4 OpenAI Codex

Codex (OpenAI's coding agent) reads MCP configuration from ~/.codex/config.toml, shared between the Codex CLI and IDE extension. Add:

[mcp_servers.lightbringer]
url = "https://mcp.lightbringer.com/mcp"
default_tools_approval_mode = "prompt"


Then authenticate:

codex mcp login lightbringer

Your browser opens Lightbringer's sign-in — sign in with email, Google, or Microsoft, approve the requested access, and select your organisation. Inside the Codex TUI, /mcp shows the server's status, and "Who am I on Lightbringer?" confirms the connection.

The default_tools_approval_mode = "prompt" line makes Codex ask before each Lightbringer action; we recommend it because the connector can create and submit real disclosures.

Where it works

• Claude (web & desktop)
  How to add: Settings → Connectors → Add custom connector · Sign-in: OAuth in browser
• Claude Code
  How to add: claude mcp add, then /mcp → Authenticate · Sign-in: OAuth in browser
• ChatGPT
  How to add: Settings → Connectors → Create (Developer mode) · Sign-in: OAuth in browser
• OpenAI Codex
  How to add: config.toml, then codex mcp login · Sign-in: OAuth in browser

Any other MCP-capable client that supports remote streamable-HTTP servers with OAuth (Cursor, Windsurf, and others) can connect the same way using the endpoint above.

Permissions: whichever client you use, the connection grants read and write across your innovation disclosures, reviews, comments, documents, and related patent-workflow data — all within the organisation you selected at sign-in.

2. What you can do

Once connected, just talk to your AI assistant in plain language about your patents and inventions — it will take the right actions on Lightbringer for you. The examples below say "Claude," but the same prompts work in ChatGPT, Codex, or any other connected client. Anything that only reads your data runs freely; anything that changes data (creating, updating, submitting, commenting) should be confirmed with you first — see Changes are confirmed under Tips for how each client handles this.

And because these assistants can use all your connected tools in one conversation, they can pull in the engineering work you've already done — code, design docs, tickets, chat — and turn it into Lightbringer inventions. Which tools are available depends on what you've connected in that client (Claude, ChatGPT, and Codex each have their own connections). That combination (see Turn your engineering work into invention disclosures below) is one of the most valuable ways to use this connector.

Explore your portfolio

"List my Lightbringer inventions."

"Search our portfolio for anything related to adaptive filtration and summarize what we already have."

"Open invention LB-104 and give me a plain-English summary of the problem and the solution."

Claude can list and search your inventions, applications, and reports, and read any document back to you as a clear summary. This is the usual starting point — it's how you find the specific invention or review you want to work on.

Turn your engineering work into invention disclosures

This is often the fastest way to capture IP. If you've connected other tools to Claude — code and pull requests in GitHub, design docs and specs in Google Drive, discussions in Slack, tickets in Jira or Linear, wikis in Confluence or Notion, or files on disk — Claude can read that engineering context and capture the inventive parts straight into Lightbringer, instead of you re-writing everything by hand.

"Read the design doc in our 'Project Atlas' Drive folder and the linked GitHub PR, then draft a Lightbringer disclosure capturing what's novel."

"Summarize the last two weeks of our #firmware Slack channel and the related Jira epic, and turn the inventive parts into a draft disclosure."

"Here are my engineering notes and test results — pull out the distinct patentable ideas and start a disclosure in Lightbringer for each one."

Claude gathers the material from your other connectors, identifies what's potentially novel, and assembles it into Lightbringer's disclosure structure — then you refine and create it exactly as below. It also makes capture routine: point Claude at the work as you finish it, and ideas land in Lightbringer before they're forgotten.

Availability depends on which connectors you've enabled — Claude can only read tools you've connected and authorized in your Claude account.

Draft a new invention disclosure

"I want to file a new invention. Walk me through what Lightbringer needs and interview me for the details."

"Draft the disclosure from my answers, check it for problems, and fix anything that's missing."

"Looks good — create it in Lightbringer."

Claude follows Lightbringer's current authoring structure and assembles a submit-ready disclosure — typically a clear problem statement and its technical causes, the invention and the technical mechanism behind it, implementation details and variants, and optional prior art and illustration ideas. The raw material can come from your own answers as Claude interviews you, from documents you paste in, or from your connected engineering tools (above) — often a mix. It checks the draft and resolves issues before creating it, and flags any remaining gaps that still need your input.

Tip: this path authors a text disclosure. If you have drawings, add them in the Lightbringer UI after the disclosure is created.

Strengthen a draft with patent-focused analysis

"Run Lightbringer's patent analysis on invention LB-104, focus on completeness, and coach me through the gaps."

"Apply those fixes to the disclosure."

Claude can run Lightbringer's patent-focused review — for clarity, problem framing, completeness, or all of them — and turn the findings into concrete revisions. The analysis sometimes takes a few minutes; Claude will wait for the results and then walk you through what to improve, one theme at a time.

Collaborate on attorney reviews

"What reviews are waiting on me?"

"Open review #482, summarize the attorney's comments and suggested redlines, and tell me what needs a decision."

"Reply to the attorney's question on the claims comment: confirm the operating range is 40–60 °C."

"Mark the review as responded."

Claude can show the reviews you're part of, read the document under review along with each comment thread and attorney redline, post comments and replies, and record your response — so you can run a review cycle without leaving the chat.

Submit when ready

"Check invention LB-104 one more time, and if it's clean, submit it for review."

Claude verifies the disclosure and then submits it. Because submitting is consequential, Claude confirms with you before doing it.

3. Tips & good practices

• Bring your sources. If you've connected engineering tools (code, design docs, tickets, chat, files), ask Claude to pull from them and capture the inventive parts into Lightbringer — the richer the source material, the stronger the disclosure.
• Find it first. Most requests work best once Claude has the specific item in hand — start by asking it to list or search, whether that's your Lightbringer portfolio or the source material in your other connected tools.
• Check before you create. Ask Claude to validate a draft and resolve problems before creating or submitting; treat warnings as real gaps to fix or consciously accept.
• Be precise in reviews. Review comments attach to an exact passage of the document — let Claude quote the wording from the review rather than paraphrasing.
• Analysis can be slow. Patent feedback may run for a few minutes; it's normal for Claude to wait for the result before presenting it.
• Changes are confirmed. Creating, updating, submitting, commenting, and replying all change real data. The connector marks these actions so that clients ask before running them — Claude confirms them by default, Codex does when approval mode is set to prompt (see setup above), and ChatGPT shows its own confirmation prompts in Developer mode. If your client lets you skip confirmations, we recommend keeping them on for Lightbringer.
• Scope awareness. Everything is scoped to the organisation you chose at sign-in. To switch organisations, reconnect and select a different one. MCP access must also be enabled per organisation in Advanced Settings.
• Drawings. Disclosures authored here are text-first; attach inventor drawings in the Lightbringer UI afterward.

4. A typical end-to-end session

1. "Who am I on Lightbringer?" — confirm the connection and organisation. This works in every client and is always the right first step.
2. "List my inventions." — see what's already captured.
3. "Read the Project Atlas design doc and the linked GitHub PR, and pull out what's novel." — gather raw material from your connected engineering tools (or just answer Claude's questions / paste notes).
4. "Draft a disclosure from that, check it, and fix what's missing." — assemble a clean disclosure.
5. "Create it in Lightbringer." — confirm when prompted.
6. "Run a completeness analysis and coach me through the gaps." — strengthen it.
7. "Apply those fixes."
8. "Check it again and submit for review." — confirm when prompted.
9. Later: "Any reviews waiting? Summarize the comments, reply to the open question, and mark it responded."