A research assistant that respects the research.
Stirps began with a simple frustration: genealogy is slow not because the records are hidden, but because holding an investigation in your head is hard. Names, dates, places, contradictions and half-finished leads accumulate faster than any one person can organize them.
What Stirps is
Stirps is a software project for AI-assisted genealogical research. Its core is an MCP server: a set of structured, genealogy-specific tools that an AI agent can call to explore a family tree, study relationships, read sources and reason about evidence.
It is not a records database, and it is not a replacement for the archives, societies and volunteers that make genealogy possible. It is the layer that helps a researcher work through them methodically.
Why it exists
A general-purpose chatbot can talk convincingly about a family tree without ever checking it. That is precisely the failure mode genealogy cannot afford — a confident sentence with no record behind it, copied into a shared tree, becomes someone else’s fact within a week.
Stirps is an attempt to build the opposite: an agent whose usefulness comes from being systematic. It should be good at noticing that a baptism predates a birth, that two people may be one person, that a branch has been asserted for three generations without a single citation.
Evidence-based by construction
The product principle is short: AI-assisted, evidence-based, human-reviewed.
Every meaningful answer should be traceable to something — a record, a source, a stated assumption, or an honest admission that the evidence is missing. Where evidence conflicts, the conflict is the finding, not an inconvenience to smooth over.
- Conclusions are presented as hypotheses, with the evidence that supports and contradicts them.
- Gaps are surfaced deliberately rather than filled with plausible text.
- Anything that would change a collaborative tree is prepared as a proposal for review.
The researcher stays responsible
Genealogical judgment is a skill. It weighs the reliability of a parish clerk, the spelling habits of a region, the likelihood that a family moved for work in a particular decade. An agent can assemble the material for that judgment and can argue a case, but it should not be the one signing the conclusion.
So Stirps is being designed around an explicit review step. The agent investigates, organizes and proposes. The person reviews and approves.
Where it is going
The near-term work is the research agent itself and the MCP architecture around it, followed by integration with the official FamilySearch API — which remains subject to FamilySearch approval.
Beyond that, the intention is to support additional genealogy services and more capable research workflows, while keeping the same discipline: assist the research, never fabricate it.