OpenHeritage Public Search

OpenHeritage agent tooling

Public, cross-agent integration files for OpenHeritage, a collaborative genealogy and cultural-heritage platform.

OpenHeritage brings together Ukraine's largest public collection of memorial, grave, and archival document data.

This repository connects compatible AI agents to OpenHeritage in three complementary ways:

• a remote, anonymous, read-only MCP server for live public-record search;
• eight portable Agent Skills with domain guidance, safe REST fallbacks, and canonical OpenHeritage links;
• a documented Personal API for user-authorized contributions with scoped bearer tokens.

No API key is required for public search.

Ukrainian heritage data at scale

• Memorials and graves: discover memorial profiles, grave sites, cemetery records, cemetery photographs, map locations, and transcriptions.
• Archival documents: search repositories and sources, retrieve digitized documents by page or original file, and use XML, transcriptions, parsing, or OCR to answer questions with provenance.

These records are connected through canonical OpenHeritage pages, so agents can move from a discovery lead to the relevant cemetery, memorial, source, or document without treating search results as proof.

MCP endpoint

https://openheritage.online/mcp

The endpoint uses MCP Streamable HTTP and exposes:

• search_records
• search_memorials
• search_cemeteries
• search_sources
• search_documents
• search_repositories
• search_places
• search_authors
• search_collections

The MCP server also exposes anonymous, read-only resources for selected public sources, documents and files, repositories, collections, canonical places, memorials, cemeteries, and Source classification tags. MCP never accepts a Personal API token.

Use the root .mcp.json with clients that support project or plugin MCP configuration.

Personal API and imports

Use the Personal API only for user-authorized contributions:

• interactive reference: https://openheritage.online/api-docs;
• OpenAPI 3 document: https://openheritage.online/api/openapi/v1.json.

Personal API tokens are bearer tokens and always include api:read. A generic catalog-only Source import uses api:sources; adding SourceDocuments, original files, page images, or PAGE XML also uses api:documents. A Source is valid without documents when no digital representation is available. The openheritage-source-import skill explains repository references, Authors, classification tags, coverage dates and places, and optional document ingestion.

Reviewing and correcting existing Sources and canonical places for a member uses api:sources, plus api:places for place changes, and never api:destructive. The openheritage-catalog-review skill explains which records a token may change, how to pick and order the Sources that need work, the evidence each correction needs, and why process notes never go into the records.

A classified newspaper import additionally uses api:authors to create or update the canonical newspaper organization authority, api:sources to create issue Sources and configure the automated Collection, and api:documents to create the issue document and upload ordered page images and PAGE XML. Send the token only in the Authorization: Bearer ... header, never to MCP or in a URL.

The openheritage-newspaper-import skill contains the complete issue workflow. It resolves the environment-specific UUID for stable Source taxonomy code record-kind-newspaper, creates or reuses the newspaper organization Author, configures an Author-driven Collection grouped by publication year, records the holding repository and publication place when known, creates each issue as a publication, and uploads its ordered document pages and PAGE XML. Standalone newspaper clipping PhotoAssets remain part of openheritage-photos.

Add OpenHeritage to your agent / Додайте OpenHeritage до свого агента

English

Use the MCP connection for live, public OpenHeritage search. Add the skills as well when you want your agent to receive focused guidance for memorials, graves, archival documents, historical photos, and genealogy research.

Codex

codex mcp add openheritage --url https://openheritage.online/mcp

npx skills add OpenHeritageOnline/agent-tooling

Gemini CLI

gemini mcp add --scope user --transport http openheritage https://openheritage.online/mcp

gemini skills install https://github.com/OpenHeritageOnline/agent-tooling

Claude Code — install the complete plugin, which includes both the MCP server and skills:

claude plugin marketplace add OpenHeritageOnline/agent-tooling

claude plugin install agent-tooling@openheritage

If you only need live search in Claude Code, use this instead:

claude mcp add --transport http --scope user openheritage https://openheritage.online/mcp

Start a new chat after installation. In an open Claude Code session, run /reload-plugins after installing the plugin. Do not install both the Claude plugin and its separate MCP entry unless you intentionally want duplicate configuration.

Українською

Підключіть MCP, щоб агент міг шукати публічні записи OpenHeritage у реальному часі. Додайте також навички, якщо хочете, щоб агент отримав спеціальні інструкції для роботи з меморіалами, могилами, архівними документами, історичними фотографіями та генеалогічними дослідженнями.

Codex

codex mcp add openheritage --url https://openheritage.online/mcp

npx skills add OpenHeritageOnline/agent-tooling

Gemini CLI

gemini mcp add --scope user --transport http openheritage https://openheritage.online/mcp

gemini skills install https://github.com/OpenHeritageOnline/agent-tooling

Claude Code — встановіть повний плагін: він містить і MCP-сервер, і навички.

claude plugin marketplace add OpenHeritageOnline/agent-tooling

claude plugin install agent-tooling@openheritage

Якщо в Claude Code потрібен лише пошук, використайте натомість:

claude mcp add --transport http --scope user openheritage https://openheritage.online/mcp

Після встановлення почніть новий чат. Якщо ви встановили плагін у вже відкритій сесії Claude Code, виконайте /reload-plugins. Не встановлюйте одночасно плагін Claude та окремий запис MCP, якщо навмисно не хочете дублювати конфігурацію.

Claude Code

Add the public OpenHeritage marketplace, then install the complete plugin:

claude plugin marketplace add OpenHeritageOnline/agent-tooling

claude plugin install agent-tooling@openheritage

Connect only the MCP server:

claude mcp add --transport http --scope user openheritage https://openheritage.online/mcp

To load the complete plugin from a checkout:

claude --plugin-dir /path/to/agent-tooling

The repository includes a Claude Code manifest at .claude-plugin/plugin.json.

Codex and other Agent Skills clients

Install the portable skills from GitHub with a compatible Agent Skills installer:

npx skills add OpenHeritageOnline/agent-tooling

The repository also includes a Codex manifest at .codex-plugin/plugin.json. Plugin-aware hosts can load the repository checkout directly and use its bundled .mcp.json.

Every Agent Skills-compatible host can also discover the eight published skills from the OpenHeritage Agent Skills index. For an MCP host without plugin support, add the openheritage entry from .mcp.json to its user or project MCP configuration.

Skills

The same versioned skill documents are also published from the OpenHeritage website:

• Agent Skills discovery index
• Umbrella OpenHeritage skill

MCP Registry

server.json publishes the remote server as io.github.OpenHeritageOnline/public-search in the official MCP Registry. The GitHub Actions workflow publishes it when an mcp-v* tag is pushed:

git tag mcp-v2.11.3

git push origin mcp-v2.11.3

The workflow uses GitHub OIDC, so it requires no stored token or domain-verification private key. Registry versions are immutable: bump server.json before creating a later tag.

Publishing is restricted to protected release refs and the mcp-registry-production GitHub Environment. The publisher is downloaded from an exact release, verified by SHA256 and Sigstore identity, and checked for the expected version and source commit before the OIDC-capable job can run. See the required repository controls before creating a release tag.

Safety

• Treat search results as discovery leads, not proof of identity or family relationship.
• Respect record visibility and copyright restrictions.
• Do not crawl, bulk-enumerate, or collect profile/contact data.
• Prefer canonical OpenHeritage pages when sharing results with a user.
• Use authenticated or mutating REST workflows only when the user explicitly requests them.
• Never print, persist, or forward a Personal API token; request only the scopes needed for the authorized operation.

Source of truth

The production skill documents are maintained in the main OpenHeritage application repository under sources/FrontendServer/AgentSkills/. Keep the copies in skills/ byte-for-byte synchronized when releasing a new version.