Skip to content
defer

One MCP. Infinite surface.

Your team shapes the MCP. Every tool. Every schema. Every prompt. Every trigger. Whatever your team writes, every agent reads.

session 01 · knowledge inheritance
⚙ tooldefer.brief()

session start

└
playbook loaded

ready in 38ms · last reshape 2h ago by laura

›fix the refresh token race in src/lib/auth.ts.

⚙ tooldefer.recall({ query: "auth refresh token race" })

non-trivial action · checking playbook first

└
matched #864 · double-CAS on token_version, not mutex. mutex deadlocked at 14k concurrent in prod.rule

rule · authored by maya · 6 months ago · recall x9

└
matched #871 · supersedes #842 (the mutex approach). lineage: tried mutex → prod deadlock → moved to CAS.decision

decision · authored by maya · 6 months ago

◆ claude

Maya's reasoning is in the playbook. She tried mutex first, hit deadlocks at 14k concurrent sessions in prod, and landed on double-CAS with token_version. I'll add the version-mismatch retry, no lock.

› Type to continue...

Conventions. Conflicts. Onboarding. Or something only your team needs.

Defer doesn't prescribe a use case. The same MCP that captures your agent's corrections can carry blocked files, onboarding briefs, decision lineage, code conflict flags. Or whatever your team decides is worth remembering.

knowledge inheritance

Maya leaves Friday. Her reasoning on auth, payments, and the deploy pipeline stays with every agent that touches those files.

decision lineage

Every architectural call carries its supersedes. The arc from "we tried X" to "we landed on Z" stays attached to the files.

replace claude.md

Stop hand-maintaining one CLAUDE.md per project. Defer lives across every tool, updates in minutes.

onboarding

New hire opens Cursor. Defer briefs them on the stack, the gotchas, and the recent ADRs.

code conflicts

Two teams hit the same file in the same week. Defer surfaces both views before either ships.

bug flags

Mark unstable code. The next agent that opens that file sees the flag before suggesting a change.

playbook A/B tests

Fork the playbook into two shapes. Different corpus, different prompts, different triggers. After two weeks: which one ships cleaner code, which one developers actually use.

employee metrics

Per-team recall counts, contribution patterns, drift cadence. Visibility without surveillance.

Find the rules your code stopped following.

Paste the prompt into any agent session. The agent reads your repo's convention docs (any combination of AGENTS.md, CLAUDE.md, .cursorrules, .windsurfrules), scans the code for rules that have drifted, filters out the rule-definitions that aren't really violations, and ranks the top 5. Runs entirely on your machine.

Get in early.

Defer is opening to a small group of senior engineering teams before the public launch. Leave your email and we'll reach out when there's room.