search_history
Returns qualified historical candidates for a question: people, events, periods and source references that may be relevant. Each result is a lead, not evidence.
For AI agents and their users
Historical comparison for AI agents: current access status
HexMind has a private local MCP prototype for historical comparison. Inside that configured environment, an agent can search qualified historical candidates and read the original passages before offering a conditional alternate perspective with citations and limits. The prototype is not publicly available today.
Status as of 18 October 2026: private local prototype only. It is not a public MCP server, not installable from Grok or Muse, and not listed in their catalogs, and it has no public API endpoint. This page is documentation; it cannot connect an agent to HexMind.
What the prototype does
The prototype is a Model Context Protocol (MCP) server that runs locally over stdio inside a configured HexMind environment. It exposes two read-only tools:
read_history_sources
Reads the original passages behind selected candidates so the agent can check what a source actually records, in its own words, before using it.
The tools are read-only. They do not write to the historical store or take actions outside the local environment. The private source corpus is not exposed by this site; the separate worked examples quote selected published sources.
The intended workflow
State the decision and timeframe
Name the choice, the people or organisations involved, the period of interest and any constraint that would rule an option out.
Search for qualified candidates
Call
search_historyto collect leads. Treat every result as unverified until the underlying passage has been read.Read the original passages
Call
read_history_sourcesfor the candidates that matter. Record what the source states, what it leaves unclear, and where the source is offering an interpretation.
Then, and only then, offer a perspective
The agent can present a conditional alternate view: “If these conditions match, this historical case suggests…; the comparison weakens when…”. It should keep three things separate: what the source records, the interpretation drawn from it, and the present-day analogy. It should cite the passages it read and state the limits of the comparison.
Candidates are not evidence. A search result alone does not support a factual claim, and an unread candidate must not be presented as a historical finding.
One concrete question
Question an agent could work on: “A small manufacturer’s sales have fallen for three months. It is considering a broad price cut on its current products. What historical business decisions are comparable, what did the original published accounts record about them, and under what conditions would the comparison apply to this situation?”
The intended output
- A short list of candidate cases, each labelled as a lead with its source reference and any time stated in the record.
- For each case the agent chooses to use, the original passage it read, a citation, and a plain statement of what the passage records.
- A conditional alternate perspective: the condition under which the historical case is relevant, the trade-off it exposes, and the points where the analogy breaks down.
- Explicit limits: differences in market, scale, regulation and available information, plus anything the sources do not establish.
- A clear separation between historical fact, the agent’s interpretation and any suggestion for the present decision.
This describes the shape of a supported answer. It is not a result produced for a customer, and it is not a testimonial or a claim about outcomes.
What this is not
- It is not a public MCP server. The adapter is stdio-only and works in a configured local HexMind environment.
- It is not installable from Grok or Muse. Neither lists a HexMind connector or integration.
- It has no public API endpoint. There is no URL an agent can call to search HexMind history.
- An agent cannot connect automatically from this page. There is no install button, connector handshake or credential flow here.
- It does not expose the private historical source store through this website.
- It does not make present-day business decisions or replace the judgement of the person who owns the decision.
A user or administrator would have to connect a future approved or public connector. No such connector is available today.
Access today
Access is arranged manually. If you want to discuss a guided pilot for historical comparison, write to support@hexmind.ai and describe the question, the kind of source material you expect to rely on, and who would review the output. Do not send confidential business records in a first message.
The business-decision email pilot described on the HexMind homepage is a separate, human-guided service. It is not the MCP prototype, and this page does not enroll anyone in it.