Context broker for
AI coding agents

Manages your agent's context window so it reads only what matters. 164x fewer tokens on large codebases.

$ npm install -g packet28

Works with your tools

Auto-detects your runtime. One command to configure MCP servers.

Claude Code
Cursor
Codex
GitHub Copilot
Windsurf

Three steps to focused context

01

Setup

Auto-detects runtimes, configures MCP servers, starts the daemon.

$ packet28 setup --root .
02

Agent calls MCP

Your AI agent calls get_context with an action: plan, inspect, edit.

get_context({ action: "plan" })
03

Focused context

Returns only what matters — repo anchors, focus sets, coverage data.

~849 tokens (not ~139,000)

Everything your agent needs

01 MAP

Repo Mapping

Builds a structural map of your codebase — files, symbols, dependencies — so the agent navigates directly to what matters.

02 COV

Coverage-Aware

Ingests test coverage data to show which lines are tested vs untested. Agents know exactly where to add tests.

03 BUD

Token Budgeting

Each response stays within a configurable token budget. No more blowing through context windows on a single file read.

04 FOC

Focus Sets

Agents mark files and symbols as "in focus." Packet28 tracks what the agent cares about across tool calls.

05 DMN

Daemon State

The packet28d daemon persists state across agent sessions. Start, stop, resume — your context survives restarts.

06 MRT

Multi-Runtime

One setup command detects Claude Code, Cursor, Codex, and more. Configures MCP servers for each automatically.


The numbers

Tested on Apache Commons Lang — 534 Java files, 12MB.

WITHOUT PACKET28
139,000 tokens
WITH PACKET28
849
164x
fewer tokens consumed
Task: "Understand StringUtils.wrap and fix null delimiter handling"
Naive agent reads entire 9,243-line files. Packet28 serves ~40 relevant lines + anchors.

Up and running in 30 seconds

TERMINAL
# install globally
$ npm install -g packet28
# set up in your project
$ packet28 setup --root .
Detected runtimes: Claude Code, Cursor
Setup complete. MCP servers configured.

How the pieces fit


Built by

Utsav Sharma
Founder
linkedin →