Skip to directory

About Semilattice MCP

A hosted service that simulates how defined audience models react to product concepts, decisions, messaging, and user journeys, supporting content testing, feature prioritisation, and UX research.

What you can use it for

  • Validate product concepts with simulated audience models.
  • Test copy and messaging options.
  • Explore user frustrations and attitudes.
  • Simulate user journeys to find conversion or usability issues.
Connection and setup
Hosting
Remote
Transport
Streamable HTTP
Endpoint
https://mcp.semilattice.ai
Authentication
OAuth browser authorization through a Semilattice account.

Requirements

  • Semilattice account
  • Client supporting remote MCP over Streamable HTTP with OAuth
  • Some clients require a paid subscription: publisher docs specify paid plans for ChatGPT web and Codex and a paid Claude subscription for Claude Desktop/web.

Setup steps

  1. Add remote Streamable HTTP server https://mcp.semilattice.ai in a supported client.
  2. Start the client OAuth flow and create or sign in to the Semilattice account using Google email.
  3. Enable the connector for the current conversation where the client requires it.
  4. Start with list_user_models or list_simulation_schema. Simulations may run from minutes to around half an hour; use run_simulation only after reviewing the plan with the user.
  5. For Codex, publisher docs require rmcp_client = true under [features], then [mcp_servers.semilattice] url = https://mcp.semilattice.ai and `codex mcp login semilattice` before launching Codex.
Tools 27

Current Semilattice tool reference lists 27 tools across context and user models, projects and rounds, simulations, and reading results. This replaces an older seven-tool inventory. Simulation runs, deletes, and sharing have material effects; run/delete are gated by explicit confirmation in the product workflow.

  • get_context

    read the context your simulations are planned against.

    01
  • list_contexts

    list the contexts you can use, yours and your organisation's.

    02
  • create_context

    write a new context with you, from what you tell it.

    03
  • list_user_models

    list the user models you can use, with their accuracy, which are your organisation's own, and for a segment, the user model it was made from.

    04
  • get_user_model_attributes

    list the attributes a user model's simulated users carry, such as age or income band, with how many hold each value, before making a segment.

    05
  • create_segment

    make a segment of a user model from the attribute values to keep.

    06
  • edit_segment

    rename a segment you made, or change its filters if no simulation has used it yet.

    07
  • create_project

    start a project, optionally with a name and a context.

    08
  • list_projects

    list your projects, newest first, to find one to add to.

    09
  • get_project

    read a project: its rounds, what each one tests, its simulations and its synthesis.

    10
  • rename_project

    rename a project.

    11
  • start_round

    start a new round in a project.

    12
  • rename_round

    rename a round.

    13
  • attach_artefact

    attach the document, page or file a round tests.

    14
  • read_artefact

    read the text of a document attached to a round, for example one uploaded in the app.

    15
  • regenerate_synthesis

    write a round's synthesis again from every simulation in it. This replaces any edits made to it.

    16
  • list_simulation_schema

    list the two kinds of simulation, their goals, and what each goal needs.

    17
  • create_simulation

    draft a simulation. Semilattice writes its questions or task in the background.

    18
  • get_simulation

    read a simulation's status, plan and results.

    19
  • update_simulation_plan

    change a planned simulation: edit, add or delete questions, or change a journey's task.

    20
  • run_simulation

    run a planned simulation, after you have confirmed it.

    21
  • list_simulations

    list past simulations, to build on what you already know.

    22
  • delete_simulation

    delete a simulation, after you have confirmed it.

    23
  • share_simulation

    make a simulation's results viewable by anyone with the link.

    24
  • read_respondent_answers

    read how individual simulated users answered.

    25
  • read_journey_session

    read one simulated user's journey in full, step by step.

    26
  • read_planner_conversation

    read what you discussed with the agent in the app on a project, so your agent knows what was decided there.

    27
Limitations and notes
  • Simulation runs can take minutes to around half an hour. Current official docs say simulations are run only after confirmation and deletion likewise requires confirmation. Sharing a simulation makes its results viewable to anyone with the link.
Original sources
  • official complete tool reference

    Current machine-readable tools page enumerates 27 tool names and concise purposes, covering more than the old seven-tool overview.

  • official product workflow

    Current introduction explains project/round workflows, simulation run confirmation, deletion confirmation, and execution duration.

  • official repository

    Maintainer repository identifies the MCP implementation and current endpoint.

Documentation checked 2026-10-28T14:06:33.000Z. Connection has not been runtime tested.