Carrying Context Between AI Tools Without Losing the Thread
I kept hitting limits in one AI assistant and starting cold in the next. OmniContext is the local bridge I built so my project state travels with me instead of getting re-explained every time.
Here's a workflow most people who code with AI will recognize. You're deep in a session with Cursor, you hit a usage limit, and you switch to Claude on the web to keep going. Except now you're pasting in files, re-describing what you changed, and reminding it what you were even trying to do. The work didn't move — your context just fell on the floor.
OmniContext is the local handoff tool I built for this workflow. Its CLI launches a foreground dashboard that reads Git state on initial load or manual refresh and formats a prompt for the next tool. It does not continuously monitor or persist active sessions.
Two ways to hand off
There are really two kinds of AI tools, and they need different bridges.
For web assistants like ChatGPT, Claude, and Perplexity, the bridge is a document. OmniContext's dashboard formats architecture, the current branch, changed files, the full uncommitted diff, and conversation you supply into a Markdown handoff prompt. It does not compress the diff or enforce a token budget. Review and trim large handoffs before pasting them.
For IDEs that speak the Model Context Protocol, including Cursor and Claude Code, a separate MCP server exposes the context://current resource. Its task tools currently operate on an in-memory demonstration list, not your persistent session history; changes disappear when the server restarts.
# Install from GitHub; the npm package with this name is unrelated.
npm install -g github:neerajbhargav/omni-context
cd /path/to/your-project
omni
# Configure omni-context-mcp separately as your IDE's MCP command.What the current handoff does and does not do
The shipped dashboard makes the inputs explicit rather than summarizing them: Git state, the full diff, architecture, and supplied conversation. That is useful for inspecting what the next assistant will receive, but it is not automatic compression or relevance ranking. Token-budget enforcement, continuous monitoring, and persistent task tracking remain outside this implementation.
The bet underneath it
I built this because I don't think any one AI tool wins, and I don't want my context locked inside whichever one I happened to start in. MCP is quietly becoming the standard that makes that portability real — so OmniContext leans on it where it can, and falls back to plain Markdown everywhere else. Your context should belong to you, not to a vendor's session.