A genealogy server for the agent ecosystem.
Stirps is built as a Model Context Protocol server. An MCP-compatible agent connects to it and gains a set of genealogy-specific tools — typed operations with clear semantics, instead of a generic search box and a hope.
- 01AI agentAny MCP-compatible client
- 02Stirps MCP serverTools, research state, guardrails
- 03Genealogy toolsPeople, relationships, evidence, planning
- 04Authorized servicesUser-authorized genealogy APIs
Why specialized tools
Genealogy has real semantics. A person is not a row; a relationship has a direction and a degree of certainty; a date can be exact, estimated, or asserted by a source that contradicts another source. Flattening that into free-text search throws away precisely the structure the research depends on.
Specialized tools let the agent be held to the shape of the domain. It can ask for the parents of a person and receive an answer that carries its own uncertainty, rather than a paragraph it then has to interpret.
Tool families being designed
The server is organized around what a researcher actually does, not around API endpoints.
- Exploration — navigate ancestors, descendants and collateral lines.
- Analysis — examine dates, places, events and relationship consistency.
- Evidence — organize sources, citations, documents and memories.
- Detection — surface missing information, conflicts and possible duplicates.
- Planning — break a question into a multi-step investigation and track its state.
- Explanation — report what supports or contradicts a conclusion.
- Contribution — prepare a reviewed change for a user to approve.
Compatible agents
Development is happening first against OpenClaw, but the server targets the protocol rather than a single client. Any MCP-compatible agent should be able to connect once the server is released.
Where a client offers richer affordances — structured approval prompts, for example — Stirps will use them. Where it does not, the review step still exists, expressed in conversation.
Guardrails, not suggestions
The review requirement belongs in the server, not in a prompt. Tools that can change external data are being designed as explicit two-phase operations: prepare, then submit with approval. A tool that can only read says so in its contract.
Developer roadmap
In order, and without dates we cannot commit to:
- Stabilize the core genealogy tool surface and the research-state model.
- Complete and submit the FamilySearch integration for approval.
- Controlled testing with a small group of researchers.
- Documentation and a public release of the server.
- Additional genealogy service integrations.
No credentials, tokens or private endpoints are published here. Configuration details will ship with the server.