This page lists everything you can ask Claude or ChatGPT to do through the
Goveda Pro MCP server, grouped by where it fits in an examination. You don't
need to call tools by name — just describe what you want ("start a project and
upload this application", "run a prior art search", "show the unity analysis")
and the assistant picks the right one. The names below are here so you know
exactly what's available.
Tools marked interactive card open a rich, clickable panel inline (in
Claude and other hosts that render them) instead of returning plain text — for
example a projects dashboard, an upload picker, or a completed search-results
panel.
Your connection is bound to the organization you signed in to — if you belong
to more than one, you pick it during sign-in — and every tool then acts in that
organization from the start.
Projects
For every project-specific tool, you can give either the Project UUID or the
exact Singapore application number as project_id. An application number works
when it identifies one Project you can read. If it matches several of your
Projects, the assistant shows the candidate UUIDs and asks which Project to
use. When starting a project with an application number already used by a
readable Project, the assistant pauses and asks whether to continue that Project
or create a separate one; it creates a separate Project only after you expressly
choose that option.
| Tool | What it does |
|---|
create_project | Start a new examination project. Pass application_no to pull the bibliographic data straight from the Singapore patent register; omit it to supply a PF1/PF37 instead. |
list_projects | List your projects; filter by status or name. |
get_project | Open one project's details and current status. |
show_projects | Show the projects dashboard. (interactive card) |
submit_project_for_analysis | Queue the project so the system runs the recommended prior art search, the summary, and the lack-of-unity analysis. |
get_project_processing_status | Check progress of each step. Claim features and claim comparison also appear here; those results are read in the web app, not through MCP — see the Roadmap. |
show_project_processing_status | Show a live processing-status panel. (interactive card) |
set_project_completion | Mark a project completed when the examination is done, or reopen it. |
delete_project | Delete a project (project owner only). |
The application document
| Tool | What it does |
|---|
show_project_documents | Show the project's documents and upload new ones. (interactive card) |
request_document_upload | Get an upload link for a file (manual upload flow). |
add_document | Attach an uploaded file to the project and queue extraction. |
list_project_documents | List the project's documents with their status and any processing error. |
delete_project_document | Remove a document from the project. |
Reading the application
| Tool | What it does |
|---|
get_application_biblio | Bibliographic data — numbers, filing date, title, inventors, abstract, priority claims. For projects created from an application number, also applicants, IPC classifications, deemed filing date, and application type (Normal / PCT national phase / Divisional). |
get_application_sequences | The biological sequence listing (sequence patents only). |
get_application_content | Read abstract / claims / description completely by default, reusing the same sections and exact cursor until is_completed=true without reconfirming. Set completion_required=false only when you explicitly want a partial, bounded, preview, or single-page read. available=false is not a cursor state. |
search_application_content | Full-text search within the application and return the most relevant paragraphs. |
get_application_summary | The AI-generated application summary — workflow status and result — produced after you submit for analysis. Status initializing is also returned before submission; use get_project_processing_status to tell the two apart. |
Project Knowledge Base
Reference PDFs uploaded through a project's Knowledge Base are extracted once
by Goveda Pro, so an assistant can work with the resulting text without reading
the PDF through a multimodal model. MCP access is read-only: upload, rename,
retry, review, copy, download, and delete actions stay in the web app.
| Tool | What it does |
|---|
list_knowledge_sources | List the project's reference PDFs, newest first, with their processing status and page count. The list contains no document text; choose a source whose status is completed before reading it. |
search_knowledge_base | Search the extracted text of every completed reference PDF, or only selected sources. Relevance mode ranks likely matches; exact mode finds a literal phrase. Each bounded result identifies its source, original filename, PDF page, and content type. Use the returned cursor for more results, or read the selected source and page with get_knowledge_source_content. |
get_knowledge_source_content | Read page-aware Markdown from one completed reference PDF. Use an inclusive page range to narrow the context, and reuse the returned opaque cursor when has_more is true. The default is a bounded read (max_chars=6000, completion_required=false); the tool does not call or summarize with a model. |
get_knowledge_source_image | Retrieve one image from a Knowledge Base Markdown image URL, together with its source, PDF page, and caption. Call once per image. The connected assistant rechecks current project access, so an expired 24-hour browser link can still be resolved through MCP. The assistant reads the returned pixels rather than guessing from the caption. If its client cannot expose those pixels, it opens the original link through the client's browser permission flow; when that link has expired, copy the Markdown again. |
Prior art & search
| Tool | What it does |
|---|
get_recommended_prior_art | The system's recommended prior art for the application (produced automatically after you submit for analysis). Returns the shortlist by default; pass tier (most_likely, possible, or all) to widen or narrow it. |
semantic_search | Run an ad-hoc semantic prior art search once and receive a completed first page of up to 10 hits inline — no success polling. Each compact hit has publication_number, nullable title, rank, and nullable url; use get_patent(publication_number, sections) for its abstract, claims, or description. By default it searches over the application and limits results to art published on or before the application's earliest effective date (pass a later published_before to search past it, or published_after to set a floor). A free-text query can be up to 10,000 characters and cannot contain NUL. Filters accept up to 50 two-letter patent-office codes, up to 100 IPC/CPC symbols such as G06F 21/60, and patent status alive or dead. |
filter_prior_art | Synchronously post-filter either the original recommended pool or a completed user search. mode="keyword" scans at most the original top 2,000 candidates with patent-highlight-v1 expressions and preserves source order; mode="semantic" selects only from that existing pool for the supplied query_text and keeps top_k matches (default 5, maximum 50). Each mode persists the complete ordered subset and returns a compact first page plus filter_id, total, and normalized filter_conditions. Provide only the parameters for the selected mode, and omit all parameters belonging to the other mode. |
get_search_status | Check an older or otherwise in-progress search ID. A successful new semantic_search or show_semantic_search call is already complete and does not need this step. |
get_search_results | Read a historical search ID or fetch later result pages using offset and limit (1–100). Pass the optional filter_id returned by filter_prior_art to page through a persisted filtered subset; omit it for the original search. Use next_offset when has_more is true. |
show_semantic_search | Run the same one-call search as semantic_search and show its completed compact results and criteria on a card; click a hit to open its full patent card. Takes the same bounded options as semantic_search. (interactive card) |
lookup_classifications | Find authoritative CPC/IPC codes from a keyword, a symbol, or a plain-language description. |
get_application_family_citations | Get patent and non-patent-literature citations raised against this application's family. It returns independent patent and npl sections by default, so either section remains available if the other fails. Pass citation_type (all, patent, or npl) to choose sections; scope defaults to simple and can be extended for broader patent-family coverage. Each NPL citation keeps its original text and includes a link to a stated DOI or, when no DOI is available, a Google search built from the available citation details. |
Patent lookup
Scope defaults are deliberately narrow. Family sections return the
same-priority family (family_scope="simple"), and backward citations return
only what the publication itself cites (scope="patent"). This keeps results
precise, but it means the lists are shorter than a full-coverage search. When
you need the widest net — a prior art sweep, an opposition, a validity study —
ask for the extended scope explicitly: family_scope="extended" on the patent
and family lookups, and scope="extended" on the backward citation tools.
(Forward citations already default to the full family, and extended is not
a valid scope there — use scope="family".)
| Tool | What it does |
|---|
get_patent | Read requested patent sections completely by default, finishing one section per cursor chain until is_completed=true (claims before description). Set completion_required=false only when you explicitly want a partial, bounded, preview, or single-page read. Common publication and application-number formats are accepted only after they resolve to a verified patent record; a temporary lookup failure is returned as an operation error, not as “patent not found.” paragraph_numbers remains scoped. (citations use dedicated tools) |
search_patent_content | Search within one patent and get back its GoVeda URL plus the matching paragraphs, ranked, each with its paragraph number as filed — so you can locate a passage without reading the whole document, fetch it precisely with get_patent, then quote it as "D1 [0034] discloses…". Where a document numbers no paragraphs, the number comes back as null, never as invented; quote such a passage by its text. |
batch_get_patents | Read bounded rows for several patents; not complete text. Number resolution is performed as one verified batch while preserving input order and duplicate rows. Unless you explicitly request a partial or single-page read, the assistant follows each row with an independent get_patent(..., completion_required=true) chain for every requested long-text section. If the lookup is temporarily unavailable, the call returns one top-level operation error instead of per-row “not found” results. |
show_patent | Show a patent card, or a side-by-side comparison of several patents. (interactive card) |
get_patent_forward_citations | The patents that cite a given patent (forward citations), paginated. scope defaults to family, matching the web view; pass patent for this publication only. |
get_patent_backward_patent_citations | The patents a given patent cites (backward patent citations), paginated. scope defaults to patent — only what this publication itself cites. |
get_patent_backward_npl_citations | The non-patent literature a given patent cites, paginated. scope defaults to patent — only what this publication itself cites. |
Lack of Unity
| Tool | What it does |
|---|
get_lou_analysis | The lack-of-unity result — whether the claims lack unity, the reasoning, and the generated Section 4 text. |
show_lou_analysis | Show the LOU analysis card — verdict, invention groups, and Section 4 text with one-click copy. (interactive card) |
Export
| Tool | What it does |
|---|
export_prior_art_csv | Export the recommended prior art, or a completed semantic search's results, as a downloadable CSV (link expires in 1 hour). For the recommended prior art, tier (most_likely, possible, all) selects how much of it to export. |
Feedback
| Tool | What it does |
|---|
submit_feedback | Send a bug report, feature request, or general feedback about Goveda Pro. |
What's next
For web-app capabilities not yet exposed through the MCP server — and
everything else in progress — see the Roadmap.