Why Trunk Exists
Understand the motivation behind Trunk and the problem it solves.
The Problem with Traditional MCPs
Today’s model-context protocols (MCPs) are noisy and fragmented. Each new capability, like fetching a token price, adding a calendar event, or summarizing a news feed,requires manually locating a plug-in, importing its schema, wiring credentials, and handling everything from rate limits to response formats.
Agents end up dragging a backpack of bespoke dependencies just to answer a single follow-up question.
Example: A Simple Query
“What’s the current price of SOL and how does it compare to yesterday?”
Becomes four manual chores:
| Step | Pain Point |
|---|---|
| Install price API tool | Version clashes, API changes |
| Configure API keys | Secret management, rotation |
| Parse response/timestamps | Glue code, edge cases |
| Rate-limit & cache | Extra infra, stale data risk |
Result: More scaffolding than intelligence.
Get Started
Deploy a Trunk runtime and see the difference.
Explore Concepts
Learn the core ideas behind Trunk’s design.
Runtime Architecture
Execution Flow
Agents embed a tiny SDK (~x kB). Each MCP call is automatically routed:
- Public data → sent to Trunk’s batch manager and published on-chain.
- Private intents → executed locally and committed as hashed records.
No extra imports. One async response.