v1.10 · MIT licensed · No telemetry

Ship features. Not context.

CodeFire is the context engine for AI coding agents. Surgical retrieval, agent-to-agent handoffs, and a living work log — so Claude Code, Codex, Gemini, and OpenCode spend tokens on your features, not on re-reading your repo.

codefire://install-mcp?client=claude  ·  one-line MCP setup
CodeFire app — task board and project intelligence
Scroll
Context layer for —
CLAUDE CODE
GEMINI CLI
CODEX CLI
OPENCODE
+ ANY MCP-AWARE AGENT
CLAUDE CODE
GEMINI CLI
CODEX CLI
OPENCODE
+ ANY MCP-AWARE AGENT
Chapter I · The waste

Your context window
is the most expensive
thing you're paying for.

You waste tokens because you brief the agent every session. You brief the agent because it has no context. Fix the second — eliminate the first.

Without CodeFire
~/work/api
$ claude
@src/** Continue the auth refactor.
Context: we're migrating to JWT. Started in
src/auth/jwt.ts. 3 of 7 route handlers done.
Remaining: payments, webhooks, admin.
Rate limiter also needs the JWT update.
Don't touch the legacy session middleware.
// 84,210 tokens before the first answer
With CodeFire
~/work/api · codefire mcp connected
$ claude
Continue the auth refactor
codefire retrieved 8 chunks:
src/auth/jwt.ts · signTokens()
src/routes/payments.ts · handler
src/middleware/rate.ts · check()
+ 5 more, ranked by relevance
// 6,140 tokens. Same answer.
Chapter II · How it works

One layer. Three
engines.

01 — Context Engine

Semantic context,
not file dumps.

Semantic search across your functions, classes, commits, and notes. Your agent gets the 10 relevant chunks — not 100 unrelated files. Smaller context, lower bills, better answers.

02 — Agent Handoffs

Claude starts.
Codex finishes.

Hand work between Claude Code, Codex, Gemini, and OpenCode through one shared task board. Auto-labeled from:claude → to:codex. Zero copy-paste, full context preserved.

03 — The Work Log

Every decision,
every session.

Tasks, decisions, and session notes captured as the agent works. The next run — or the next agent, or the next teammate — picks up exactly where things left off.

4
CLI agents supported
Surgical
Context retrieval, not context dumps
0
Telemetry collected
MIT
Open source license
Plus the basics, done right
Integrated terminal
Run commands, watch your agent work — in-app.
Session history
Token costs, tool calls, file changes — every run, observable.
Git intelligence
Diffs, branches, and PR context the agent reads alongside you.
Pinned briefings
Architecture decisions and gotchas, always surfaced when relevant.
Chapter III · One layer, every agent

Bring your own agent.
Context is on us.

Chapter IV · Two minutes

Install in
two steps.

download codefire Step 1

Grab the latest build for your platform. macOS app is signed & notarized. Windows ships an NSIS installer; Linux comes as AppImage or .deb.

connect your agent Step 2

One-click MCP install for any supported CLI — these deep links open CodeFire and wire up your config automatically.

Or use the in-app installer — done by the time the kettle whistles. Need manual JSON config? Full setup guide.

Built by developers who got tired of paying for the same context, twice a day.