Segue: Cross-AI Context Transfer via MCP

Segue: Cross-AI Context Transfer via MCP

Segue enables seamless context migration between AI assistants

Segue provides a way to transfer working state—such as briefs, constraints, and project decisions—across different AI tools without manual copy-pasting or re-explaining. By acting as a neutral relay, Segue allows a user to save a block of context in one assistant (e.g., Claude) and retrieve it in another (e.g., Cursor or ChatGPT) using a short, three-syllable handle.

Technical Implementation via Model Context Protocol (MCP)

Segue is built on the Model Context Protocol (MCP), an open standard stewarded by the Linux Foundation that enables AI assistants to connect to external tools.

How the Transfer Process Works

  1. Connection: Users connect MCP-capable AI clients to Segue using OAuth.
  2. Saving: An AI calls the save_context function with the relevant working brief or notes. Segue returns a short, pronounceable handle (e.g., brelavo).
  3. Loading: The user provides that handle to a different AI assistant, which calls load_context("handle") to retrieve the exact state and resume work.

Supported Clients

Segue works with any client that supports MCP with streamable HTTP and OAuth, including:

  • Claude (Desktop and Code)
  • ChatGPT
  • Cursor
  • Windsurf
  • Codex

Core Design Principles

Segue is designed as a deliberate, human-scale relay rather than an automated synchronization service.

  • Explicit Transfer: No data syncs automatically. Transfers only occur when a user explicitly saves a handle in one AI and loads it in another.
  • Human-Scale Handles: Instead of long URLs, Segue uses three-syllable words that can be easily spoken aloud or typed on mobile devices, even for contexts up to 100,000 characters.
  • Vendor Neutrality: Segue does not belong to any specific AI provider, ensuring portability across current and future MCP-compliant tools.
  • Privacy and Security: Handles resolve only within the user's own account. AI clients are granted scoped permissions and short-lived tokens, which can be revoked via settings.
  • Data Portability: All saved contexts are stored as plain Markdown (.md), allowing users to browse, edit, search, and download their data to avoid vendor lock-in.

Use Cases for Context Relaying

  • Cross-Tool Workflows: Planning a project in Claude, implementing the code in Cursor, and drafting documentation in ChatGPT while maintaining a consistent brief.
  • Session Management: Saving a curated brief at the end of a day or when hitting a context window limit to avoid dragging a bloated transcript into a new session.
  • Standing Briefs: Maintaining a permanent set of project backgrounds, house styles, and preferences to bootstrap any new AI assistant instantly.

Pricing and Tiers

Segue offers a free tier and a paid Pro tier focused on storage capacity rather than feature access.

Feature Free Tier Pro Tier ($5/mo or $48/yr)
Saved Contexts 20 2,000
Connected AIs 2 Unlimited
Core Features Save, load, search, web editor, MD export All Free features + early access

According to the service, loading, editing, and exporting data remains free regardless of the account state.

Community Perspectives and Critiques

While the concept of discrete data handles for LLMs is seen by some as a solution to the lack of native state management in web-based AI apps, other users have raised concerns regarding security and utility.

Security and Sovereignty

Critics argue that sending context to a third-party relay in plain text introduces unnecessary risk.

"This is dangerous. You're sending your context to someone's relay in plain text and loading it in plain text."

Redundancy vs. Convenience

Several users pointed out that existing workflows—such as maintaining a TODO.md or AGENTS.md file within a project directory—achieve the same result without requiring a third-party service.

"Why not just ask first agent to update AGENTS.md adding context and next steps, and ask the next agent to continue from there?"

Suggestions for Improvement

Community feedback suggests that the tool would be more appealing if it were open-source or allowed users to host the context in their own storage (e.g., Google Drive) rather than on Segue's servers.

Sources