Taste
Docs

Use the same design context across AI tools

Keep Claude Code, Codex, and Cursor working from the same references and decisions without assuming they share a conversation.

Bring the references and the decision

You might explore a screen in Claude Code, implement part of it in Codex, and return to Cursor for a small fix. Each authorized Taste connection can request permitted context from your account. That gives the tools a common place to find references and notes.

The connection does not transfer your other tool’s chat history, local files, uncommitted changes, or reasoning. “Continue what we discussed” needs an accompanying handoff. Keep two things explicit: the Taste collection that informed the work, and the decisions you want the next tool to preserve. The current project still needs to be available to that tool.

A small handoff is enough if it contains the choices

Imagine an account-settings project. In Claude Code, you studied a quiet navigation reference and a compact form reference. You chose to keep the existing sidebar, group notification options by purpose, and use the project’s current switches. No implementation has happened yet. Codex cannot infer that last detail from the references.

Write those decisions into a short note you deliberately provide with the next task. Include the collection link, the target screen or file, what has actually changed, and what remains open. Name a rejected direction when it would otherwise be tempting to repeat it: for example, replacing the sidebar with tabs. This is your handoff, not an automatic Taste chat export.

Try this prompt

Collection: [paste the Taste collection URL] Target: [name the settings screen and its file or route] Decisions: keep the sidebar, group notification options by purpose, and reuse the existing switches. Replacing the sidebar with tabs was rejected because other settings pages use it. Implementation status: no code changed. Still to decide: whether email and push settings belong together or in separate groups. Acceptance checks: keep every existing option available; verify keyboard access and the narrow layout. Read this collection through Taste and inspect the current project. Confirm the references and decisions you can access, identify missing context, and propose the implementation.

Verify one recognizable detail in each tool

Use the same Taste account and identify the same collection in each client. A precise collection link helps when names are similar. Ask the next tool to report a reference and summarize a note you recognize. A successful sign-in alone does not establish that it found the collection you meant.

Tools can differ in the capabilities their connection exposes and in whether they can view returned images. A screenshot URL is not evidence that the agent inspected the screenshot. If one tool can read your note but cannot see the image, give it work the note supports, or provide the missing image through that client. Do not ask it to fill in the visual details from memory.

Try this prompt

Read my Account settings collection through the live Taste connection. Name one reference, summarize one of my available notes, and state whether you inspected its image. If the collection, note, or image is unavailable, identify the missing part. Do not substitute a local skill export without telling me.

A new read and an old conversation are different things

Suppose you change a reference note from “try tabs” to “keep the sidebar.” The earlier conversation still contains the old instruction. Ask the agent to reread the changed note and revise its plan. It should not need to reload an unchanged collection after every small code edit.

A local Taste skill export is a snapshot. It may be useful when live tools are unavailable, but it does not establish that the current collection was read. A copied brief is also a snapshot of your decisions. When those documents disagree with a newer note, state which direction applies instead of expecting the agent to guess from whichever text it encounters last.

Keep the profile’s age separate from the library’s age

MCP reads the latest stored Taste Profile when the tool requests it. Saving another reference changes the library; it does not by itself regenerate that profile. Generating or regenerating a profile and asking the tool to fetch it are separate steps.

For the settings example, the collection notes can direct the immediate work even if the profile still reflects broader preferences. Say that this project needs the existing sidebar and components. You do not need to regenerate a personal profile merely to explain one screen’s constraints.

Compare the implementation to the same brief

Before switching again, record what was implemented, what was checked, and what still needs attention. “Grouped notification settings; keyboard review still needed” is more useful than “mostly done.” Make the current code available through your normal project workflow; a Taste collection does not move it between workspaces.

Review each tool’s output against the same task and brief. The goal is consistent decisions, not identical generated code. If the implementation changes your mind, update the handoff and the relevant reference note deliberately. A suggestion in one conversation does not become an agreed project rule just because another tool can read the same library.