mossbear
how it workspricingdocs
sign in

stop re-teaching your agents.

Mossbear is the context layer above your repos. Capture rules, skills, and notes once — from any device — and every agent you run works from them: Claude Code, Codex, Cursor, local or remote. Then every run gets graded, so you know which context actually works.
npm install -g mossbear

works where your agents work

Claude Code
Codex
Cursor
Copilot
Gemini CLI
anything that reads AGENTS.md
Built on the AGENTS.md standard — 60,000+ projects already use it.

sound familiar?

"I maintain the same rules in three files."

CLAUDE.md, AGENTS.md, .cursorrules — edit one, forget two. Your conventions drift apart per tool.

"My cloud agent doesn't know anything."

Your remote Codex agent spins up in a fresh container while your local Claude Code knows all your conventions. Same project, split brain.

"Half my context lives outside the repo."

Notes, specs, references — scattered across apps and devices where no agent can reach them. So you paste the same background into every session.

"I have no idea if my rules even work."

You write instructions, the agent does… something. The only feedback loop is your own frustration.

how it works

$ mossbear initdetecting platforms… found Claude Code, Cursorimporting CLAUDE.md, .cursorrules…done — your docs are live.
1

mossbear init

auto-detects your platforms, imports the rules and skills you already have. You don't start from scratch — you bring your CLAUDE.md with you.
AGENTS.md
2

work like you already do

Mossbear projects your docs to every connected platform's native format and keeps them in sync — including the repos your remote agents clone. Agents pull scoped context on demand through the built-in MCP server, and corrections you make in conversation become draft guides automatically.
✓followed the API error-handling guide
?no guide matched this change
✗skipped the test-first convention
3

see what worked

Every run lands in one graded list across all your agents. Each guide carries a performance score. Refine the ones that underperform, keep the ones that pull weight.
canonical

one source of truth, every platform

A canonical AGENTS.md plus a `.agents/` directory, projected automatically to every platform's native format. Canonical items, not per-tool copies — edit once and every platform follows, with bidirectional import and file watching to keep it that way.
sync
cross-repo

context that lives above your repos

Capture guides, notes, and references from any device into your inbox. Agents pull exactly the context a task needs — `mossbear pull` or the built-in MCP server — and push polished results back as canonical assets. Your best context stops being trapped in one repo.
repo a
→
repo b
measurable

context with a feedback loop

Every run is graded against your guides — aligned, unguided, or misaligned. Per-guide performance scores show which instructions actually work, so evidence travels with your context: not just what a skill contains, but what it did. And guides write themselves from your own corrections.
✓error handling guide — 92% aligned
✓commit message guide — 88% aligned
?test coverage guide — needs review
✗naming convention guide — 41% aligned
scale & planning

plan once, run many

Planning docs and specs are managed alongside your skills, so parallel local and remote agents stay coordinated. One dashboard shows what every agent did across every platform.

your code stays on your machine

Mossbear evaluates every session locally. Only verdicts and metadata sync to your dashboard — your code and diffs never leave your device.

stays on your device

Never uploaded, never stored on our servers.
Raw conversation transcripts (read on your machine to count tokens)
Your file contents
Code diffs
Shell commands you run
The local verdict queue

syncs to your account

Enough for the dashboard to grade your alignment — nothing more.
Verdicts & action summaries
Guide suggestions
Feedback signals
Token counts & run usage
Skill load-overhead token counts
The CLI version making the sync request
Which agent sessions have finished
Which rule files this repo has on disk (paths and hashes, never contents)
This isn't a policy you have to trust — it's where the code runs. The CLI is open source, so you can read every line that touches your files.
local-first
open-source CLI
your keys

pricing

Free

$0multi-platform and synced from day one — everything you need to feel the loop click.
✓unlimited platform connections — Claude Code today, plus anything that reads AGENTS.md
✓cross-platform sync — one set of docs, every agent
✓full AI evaluation & grading, every run
✓up to 5 active guides
✓7-day activity history

Pro

coming soonthe depth features for power users, launching soon — sign up free today and you'll be the first to know.
✓spend tracking — real dollar cost per run, by model and by skill
✓extended history — well beyond the free tier's 7 days
✓secure credential store — bring your own provider keys, encrypted
✓self-optimizing guides — AI-suggested improvements drawn from your own corrections
✓performance trends & guide A/B testing
no card needed — upgrade when Pro launches.
no credit card for free; the CLI is open source either way.

teach your agents once. every repo remembers.

npm install -g mossbear
docspricingtermsprivacy
named for the mossbear — the tardigrade, which dries into a tun, waits out the gap, and walks again the moment water finds it.