Happy slice of mcp pizzaMCP.pizza
waggle

waggle

MCP.Pizza Chef: modiqo

When one AI assistant hands work off to another, it usually has to paste an entire document into the conversation, which eats up space and slows everything down. Waggle replaces that with a small stand-in reference pointing back to the original file, so each assistant pulls only the part it actually needs. It also keeps a record of who looked at what, and if the file changes, every assistant holding that reference gets the update automatically. It runs as one program on your computer and connects to apps like Cursor with a single line of setup — no account required.

Coding
Data
Files/PDF

Use This MCP server To

Hand off a long document to another AI assistant without resending it Track exactly which parts of a file an AI assistant read Give each AI assistant only the slice of a file it needs Fix a mistake in a shared file and update every assistant instantly Share files between AI assistants running on different computers See a record of every time an assistant opened your file

README

the waggle mark: a figure-eight dance with the waggle run as an arrow waggle

Not a path. Not a URL. A handoff that answers back.

Locations are dumb — no per-agent shaping, no receipts, no way to fix them once sent. waggle's 30-byte token resolves into each agent's own view, counts every read, and propagates a correction to every holder. A path can't do that; a URL needs a server; this needs neither.

modiqo%2Fwaggle | Trendshift

modiqo%2Fwaggle | Trendshift

The problem · How it works · Install · Harness setup · By file type · Reach · Switchboard · The essay · The paper

The handoff, before and after: pasting the whole artifact to every subagent, versus handing off a 30-byte token that each consumer resolves into its own projection

The problem

We are entering the world of agent harnesses: Claude Code orchestrators fanning out subagents, Codex sessions delegating in parallel, cross-vendor agents discovering each other over open protocols. And every one of these handoffs, today, works the same way: forward the context and hope.

The costs are measured, not hypothetical. Multi-agent systems consume ~15× the tokens of a chat session — the overhead attributed by the vendor itself to "duplicating context across agents… and summarizing results for handoffs," whose one-line summary is "each handoff loses context." Roughly 37% of multi-agent failures trace to exactly this seam.

Waggle's competitor is not another protocol. It is "Here's /tmp/analysis.md. Use it." — and that instinct is correct: a path is a 30-byte reference, which is exactly the right size for a handoff. But a raw path has no attribution (who made this, from what), no adaptation (the small-context model gets the same 9,000 tokens as the frontier model), no lifecycle (a stale path silently serves wrong data forever), no telemetry (which subagent actually read its input? which stalled?), and no reach (it dies at the machine boundary).

Three context windows compared: a full handoff fills the window with the artifact again; a raw path is cheap but blind; a waggle token stays small and pulls back only budgeted slices through resolve, search, and read

Only the string enters the consumer's context — the artifact behind it never travels unless something fetches it. Waggle standardizes that third pattern and enforces its one hard rule by type: the token travels; the artifact never auto-expands; resolve, read, and search return only the projection or slice the consumer asked for, under byte budgets. Cheap like a path — but the reference answers back.

waggle FAQ

Can I use this to hand off a long report between two AI assistants without pasting the whole thing?
Yes — you create one small reference to the report, and each assistant that needs it pulls only the portion it actually needs, instead of the whole document being copied and pasted again.
Can I use this to check whether an AI assistant actually read a file I gave it?
Yes — every time an assistant opens the reference, that read is logged, so you can see whether it was opened and which parts were actually looked at.
Which apps does this work in?
It's built for any AI assistant tool that speaks the Model Context Protocol, including Cursor, Claude Code, and Codex-based setups, since it plugs in as a single line in your assistant's settings.
Do I need an account or an API key to use it?
No — waggle runs as one program on your own computer alongside a small local file that stores its records, so there is no sign-up or key needed to get started.
How hard is it to set up?
You add one short line to your AI assistant's configuration file, similar to installing a plugin, and it starts working — no coding required beyond that copy-and-paste step.
Will my files or documents leave my computer?
No — the reference points back to the file on your machine, and only the small piece an assistant asks for is shared, so the full document is never sent out or stored elsewhere.
Can I fix a mistake in a file after I've already shared it?
Yes — you can correct the original, and every assistant that was already given a reference to it will pick up the correction automatically the next time it looks.
Does this work if my AI assistants are on different computers or servers?
Yes — the same small reference works whether the assistants sharing it are on the same machine, on separate computers, or reached over the internet.