---
title: "How to Use Claude for Genealogy: Reading Records, Planning Searches and Connecting Your Archive"
description: "Claude can read a difficult register entry, turn your notes into a research plan and, once connected, work on your actual family archive. Here is how to ask it well, which permission settings to..."
canonical: "https://kleiobase.com/blog/how-to-use-claude-for-genealogy"
last_updated: "2026-11-02"
---

# How to Use Claude for Genealogy: Reading Records, Planning Searches and Connecting Your Archive

Claude can read a difficult register entry, turn your notes into a research plan and, once connected, work on your actual family archive. Here is how to ask it well, which permission settings to change first, and why a connector should start read-only.

Claims about Claude and KleioBase last verified 2026-10-30.


You have an Italian birth act, an *atto di nascita* from 1891, and you can make out the father's name and not much else. You upload the photo to Claude and ask it to read the entry. It gives you the date, the declarant, the witnesses and the child's name, and adds a note that one surname is hard to read. That is Claude at its best: reading a document you supplied, and saying where it is unsure.

Ask the same conversation who the child's grandparents were, and you are asking it for something that is not on the page. Claude can only guess at that.

**The short answer.** Use Claude for genealogy to read, translate and organise the records you give it, and to plan searches, but never as the source of a fact about an ancestor. Tell it what kind of document it is looking at, ask it to mark every uncertain word, and compare its reading with the image. To have it work on your own research instead of a pasted copy, add your family archive as a connector under Customize > Connectors, connect with read-only access first, set tools that change data to need approval, and turn those tools off before you use Research mode, because Anthropic says Research can call connector tools without asking.

## Claude reads the page you give it and guesses at everything else

**What it does well comes from the page.** Claude can usually identify the kind of record in front of it, explain its formula, and read most of the hand. A team led by the historian Mark Humphries at Wilfrid Laurier University tested multimodal models from Anthropic, OpenAI and Google, including Claude 3.5 Sonnet, on 18th- and 19th-century English handwriting and posted the results on arXiv in November 2024. They found character error rates of 5.7 to 7 per cent, falling to as low as 1.8 per cent when a model corrected an existing transcription. Newer models have been released since, and the study tested English handwriting, not Italian or Polish. Treat the figure as a sign that a first draft will be good. It promises nothing about your page.

**What it gets wrong comes from memory.** Without a tool connected, Claude has no access to registers, indexes or certificates. Ask it about a particular person and you get text shaped like an answer. William Walters and Esther Wilder showed how far that shape can mislead in *Scientific Reports* in 2023: of 636 citations that ChatGPT produced for short literature reviews on 42 topics, 55 per cent from GPT-3.5 and 18 per cent from GPT-4 were fabricated. They tested a different assistant on academic references, not Claude on parish registers, and we found no study that measures the second. We searched in October 2026 and found warnings but no measurements. The safe assumption is that any assistant answering from memory can invent a record, and that a fabricated citation looks the same as a real one.

Our comparison of [which AI tools can actually find your ancestors' records](/blog/which-ai-can-find-your-ancestors-records) explains why a search engine, a record hint and a chatbot each fail in a different way.

## Four habits that make Claude read a record well

**Say what the document is.** "This is a Polish Catholic baptism register entry from Galicia, written in Latin in a printed columnar form, around 1870. Transcribe each column, then translate." Context tells Claude which words to expect, and which column holds the godparents.

**Ask it to show doubt.** "Put any word you cannot read with confidence in square brackets with a question mark. Do not guess at names." Claude is more useful when it tells you which surname it is unsure about than when it hands you a clean, wrong one. Our guide to [how accurate AI handwriting transcription is](/blog/how-accurate-is-ai-handwriting-transcription) explains why names are where reading errors gather.

**Separate the reading from the meaning.** Ask first for what is written, abbreviations and all, then for an interpretation. On a US World War I draft registration card, "what does it say?" and "what does it tell me about him?" are different questions, and mixing them lets an inference slip into the transcription.

**Give it facts and forbid additions.** For a research plan or a family story, list what you know and end with "Use only these facts. If something is missing, say it is missing." Without that last instruction, an assistant can fill a gap with a plausible detail you never gave it.

All four habits share a weakness: you are pasting. Each conversation starts from what you copied in, the copy is out of date as soon as you confirm another record, and Claude cannot open the record behind a fact to check it. A connector removes that step.

## Adding your family archive to Claude as a connector

This is our product, so read this section as disclosure: we make KleioBase, an AI research companion that turns scattered historical records into one connected family archive, and it connects to Claude as a connector. Any genealogy service with an MCP server connects the same way.

MCP, the Model Context Protocol, is the open standard Claude uses to ask other services for information while it works. Once connected, Claude can search your people and records, read family trees and timelines, open the record behind an extracted fact and look at the scan itself.

**From the directory.** KleioBase is listed in Claude's connector directory. When we checked on 30 October 2026, it carried a Community label, which Anthropic's documentation uses for third-party connectors it has screened but not reviewed in depth, so Claude shows you a reminder before you connect.

1. In Claude, open **Customize > Connectors** and search the directory for KleioBase.
2. Select it, then **Connect**.
3. Claude opens a KleioBase page. Sign in, choose **Read only** or **Read and write**, and select **Allow access**.

**As a custom connector.** If you prefer, add it by address: under Customize > Connectors, select **Add**, then **Add custom connector**, name it KleioBase and enter `https://kleiobase.com/api/mcp`. Anthropic's help article on custom connectors says they work on the Free, Pro, Max, Team and Enterprise plans, with Free users limited to one. On Team and Enterprise, an Owner adds connectors for the organisation under Organization settings > Connectors and members then connect with their own account.

To use it in a chat, select the **+** button, then **Connectors**, and check that KleioBase is on for that conversation. The [Connecting AI assistants guide](/docs/ai-assistants) keeps the current steps.

## Set the permissions before the first question

Claude and KleioBase each give you controls, and Anthropic's privacy setting is a third. Set them before you start.

**Choose Read only when you connect.** A Read only connection can look up anything in your archive and change nothing. If Claude tries to change something, KleioBase answers that the connection is read-only.

**Leave tool approvals on.** Anthropic's guidance is to click "Allow always" only for a server and tool you trust to run unsupervised. Under Customize > Connectors you can also set any single tool to Blocked. Of KleioBase's 88 tools, 35 only read, and every tool that deletes, overwrites or merges is marked as destructive even when it can be undone, so Claude can tell which is which. Each request also carries a one-sentence description of what it is about to do, which Claude can show you when it asks for approval.

**Turn write tools off before Research.** Anthropic's help article says that in Research mode Claude can invoke tools from your connectors automatically without further approval, and recommends turning off tools that can make changes. For a genealogy connector, that means Research gets the lookups and none of the merges.

**Decide what Anthropic may keep.** Anthropic's update to its consumer terms in 2025 lets Free, Pro and Max users choose whether their chats are used to train models, in Privacy Settings, at any time. If you allow it, new and resumed chats are kept for up to five years; if not, the period stays at 30 days. What Claude reads from your archive becomes part of the chat, so make that choice before you connect a tree with living relatives in it. Our guide to [what not to publish about living relatives](/blog/genealogy-privacy-ethics-publishing-family-data) covers why that matters.

**Know where to cut it off.** Settings > Developer in KleioBase lists every connected assistant, its permission and when it was last used. **Revoke** stops it at once.

## Five questions a connected Claude can answer

**"Who in my tree has no death date, and what is the last record I have for each?"** Claude searches the whole workspace, not only the people you remembered.

**"Compare the names extracted from the Rossi birth act with the scan, and list anything that looks different."** It opens both and gives you a list to check against the image yourself.

**"Write a timeline of Józef Nowak's life, and name the record behind each event."** Every event is tied to a record in your archive, so each line can be checked.

**"What records should my archive have for the Nowak family but does not?"** KleioBase keeps a list of the records each person should appear in, based on where and when they lived, and Claude can turn it into a search plan.

**"Show me possible duplicate people, and for each pair compare what their records say. Do not merge anything."** Read its comparison, then decide. The instruction costs nothing, and a Read only connection would refuse the merge in any case.

## What KleioBase does with Claude, and what it does not

We are a tool in this space, so read this as disclosure.

KleioBase is an AI research companion that turns scattered historical records into one connected family archive. You upload record photos, review what was read from them and confirm the people, who are then linked across every record that names them, with duplicates and research gaps flagged. A connected Claude works on that same archive. Connecting is included on every KleioBase plan, including the free Explorer plan; uploading is free, processing a record uses credits, and your plan's people and upload limits apply. Research findings, KleioBase's suggestions about gaps and inconsistencies, start on the Researcher plan.

The [Research Companion](/docs/research-companion) inside KleioBase does this without a second account. The connector suits the times you are already working in Claude, and [the announcement](/blog/use-claude-chatgpt-with-your-family-tree) explains why we built it.

There are three honest limits.

**Claude can still be wrong about your family.** It reads the same records you do, but the conclusions it draws are its own. Ask which record a claim came from, and read that record before you accept a merge or a correction.

**Photos only.** Uploads must be JPG or PNG, one image per record. For photos on your computer, ask Claude to create an upload link: KleioBase makes a page, valid for 15 minutes, where you drop the images in. Save a multi-page PDF as separate page images first.

**It reads your archive, not anyone else's.** A connector gives Claude your research, not FamilySearch, Ancestry or an archive catalogue. Finding a record you do not have yet is still a search.

## Where to start with the record you could not read

Connect with Read only, and leave approvals on.

Pick a record you uploaded and were never sure about. Ask Claude to compare the extracted names with the scan, and to put every uncertain word in brackets.

Correct what it finds yourself, in KleioBase. Then ask what records your archive expects for the people in that record, and choose one to search for this week.

The official who wrote that *atto di nascita* in 1891 was keeping a town's civil register, with no idea that you, or a machine, would ever read it. A connected Claude helps because it can show you which entry an answer came from, and you can then read that entry yourself.

Canonical: https://kleiobase.com/blog/how-to-use-claude-for-genealogy