The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Lightbringer listing page.
Work with Lightbringer's patent service from your AI assistant. Lightbringer has qualified patent attorneys on its team providing advice, strategy assessment, novelty searches, FTO, drafting, filing and prosecution through professional engagements.
Automated feedback and professional service requests have separate lifecycles. start_innovation_feedback returns one task_id; get_task_status(task_id) returns the same status/progress/results contract. Poll while queued or running; stop at succeeded, partially_succeeded or failed, preserving successful findings and explaining per-analysis errors. request_patent_preparation returns an innovation ID/link and outcome: requested | already_requested, with no task ID. It does not confirm completed preparation or filing. There is currently no MCP endpoint for tracking professional-service milestones.
Tasks and findings expire 30 days after creation; reading does not consume them or extend retention. Use list_tasks, optionally filtered by invention_id, to recover a lost task ID in the connected organisation. Follow next_cursor even if access filtering returns an empty page; listing reports recorded status without polling. delete_task permanently removes the user’s task and findings when requested, in any execution state, with write consent. Deletion does not cancel the analysis, delete the innovation or withdraw a service request.
Version 1.1.1 supports the Lightbringer MCP innovation workflows. Public discovery verified the tool catalog and prompts on 2026-09-15; this does not establish authenticated workflow or host installation testing. Update existing installations to use the renamed tools and these three workflow skills. See the changelog for package updates. Host-directory publication is separate from the repository release; see distribution and validation.
https://mcp.lightbringer.com/mcp, with OAuth and organisation-scoped access.Registration completes when register_innovation saves an innovation description. The separate request_patent_preparation action requests patent preparation only when the user explicitly wants Lightbringer to patent that innovation. Ordinary capture never submits automatically. Registration uses the server's structured template and validates before saving in the same request. Validation errors leave nothing registered; successful saves return the ID/link and non-blocking warnings. Blocked saves are reported as pending registration. Agents cannot make payments.
You need a Lightbringer account with access to the organisation you want to connect.
Plugins are available on Claude Pro, Max, Team and Enterprise plans. To install from this repository:
https://github.com/lightbringer-patents/claude-plugin and add the marketplace.You can also upload a complete Claude plugin package through the custom-plugin upload option. Team and Enterprise owners can distribute the plugin through an organisation marketplace. Availability depends on organisation policy. See Anthropic's plugin installation guide.
If the installation summary asks you to activate changes, run /reload-plugins. Use /mcp to inspect the Lightbringer connection and start authentication if needed.
The connector uses OAuth 2.1 (Authorization Code + PKCE, S256) with Dynamic Client Registration. The server advertises its authorization server via RFC 9728 protected-resource metadata (/.well-known/oauth-protected-resource); on first use, clients prompt you to sign in to Lightbringer. Supported scopes are mcp:read and mcp:write.
Complete OAuth when prompted and select the organisation to connect. Existing record permissions apply. Confirm that the Lightbringer connector and all three skills appear in the host where you installed the plugin.
This repository is Lightbringer's own marketplace. Inclusion in Anthropic's reviewed community catalog is a separate submission; inclusion in its curated official marketplace is Anthropic's decision. Claude Code installation does not itself establish availability in every Claude chat or managed workspace. Verify the three skills and connector in each target host.
Ask naturally, or select an installed skill from the / or + menu in Claude chat and Cowork. In Claude Code, use the namespaced commands below:
| Workflow | Claude Code command | Example request |
|---|---|---|
| Capture innovations | /lightbringer:innovation-capture | “Help me identify and register innovations in the work we’ve been discussing.” |
| Request patent preparation | /lightbringer:patent-preparation | “I want Lightbringer to prepare this innovation for patent filing.” |
| Respond to reviews | /lightbringer:patent-review | “Show the reviews awaiting my response and help me act on them.” |
The skills/ tree is a verbatim mirror of agent-plugin. Make shared changes there first, then copy the entire tree here. Keep .claude-plugin/ and .mcp.json metadata separate. See the distribution guide.
Before release, validate both manifests and the skills explicitly:
Then run claude --plugin-dir . and exercise the three skills and OAuth connection. Verify installation and the workflows in each target Claude app before claiming compatibility; structural validation alone does not test authenticated behaviour.
For each package release, bump version in .claude-plugin/plugin.json and add an entry to CHANGELOG.md. Claude Code uses this explicit version to detect updates; a new commit with the same version does not update existing installations. See Anthropic's version management reference. Claude package versions can advance independently for packaging changes while the shared skills remain identical to agent-plugin.