The context layer
for AI-native development
Structure your repo, control your context budget, and feed your agents curated lessons and debugging knowledge — all through a single MCP endpoint.
Install
Two minutes to set up.
Works with whatever you already use.
Pick your client. Paste the snippet. Your agent gets every lesson, ready when it needs them.
claude mcp add maind \
--transport http \
https://mcp.maind.dev/mcp \
--header "Authorization: Bearer ${MAIND_API_KEY}"Need an API key? Get one free →
How maind works
Three moments where context matters.
Your agent needs different knowledge at different phases. maind delivers the right support at each one.
Scan
When your agent reads a prompt, maind identifies the right lessons for the task — narrowed by repo, language, and intent.
Shape
As your agent works, maind reminds it to keep the repo AI-ready: good conventions, repo-specific patterns and session boundaries that actually help.
Serve
Lessons are injected as context the agent actually reads — not buried in a sidebar. Your code shows the difference.
See it in action
Watch a real session.
maind doing its thing.
An agent starts a task, pulls lessons from maind, and writes code that reflects them. This is what adoption actually looks like.
Repo readiness
A codebase shaped for
how agents actually read.
Your agent keeps the repo AI-ready as it works — flagging structure gaps, suggesting CLAUDE.md entries, keeping docs honest. You stay in control.
Context budget
Smaller context.
Sharper reasoning.
Even with infinite context windows, smaller focused contexts cost less, run faster, and produce better reasoning. maind teaches your agent when to split, summarize, and skip.
Features
Built for how AI agents actually work.
Not another lesson search engine. A structural quality layer that gets more valuable as models get better.
Curated AI lessons
A reviewed library of workflow patterns, gotchas, and debugging knowledge — pulled into your agent at the right moment, not as one giant dump.
Repo structure guidance
maind teaches your agent to keep repos AI-ready: clear conventions, CLAUDE.md and AGENTS.md patterns, documentation that pulls its weight.
daily use
Context budget control
Cut token costs, latency, and attention noise. Your agent learns when to split sessions, when to summarize, and what to leave out.
Works with every MCP client
One endpoint, any client. Streamable HTTP for hosted use, self-hosted Docker image for enterprise on-prem deployments.
Architecture
One MCP call.
Every lesson your agent needs.
Streamable HTTP, served from edge. Sub-80ms p99. Self-hostable for enterprise.
Changelog
Shipping every week.
New lessons, features, and infrastructure improvements — tracked so you always know what changed.
Session briefings now support custom instructions
Teams can pin repo-specific lessons to be included in every session briefing. Useful for enforcing conventions across a codebase.
14 new lessons on MCP debugging patterns
Community contributions focused on stdio transport pitfalls, OAuth flow debugging, and handling connection resets. All reviewed and curated.
Sub-80ms p99 latency, now across 6 edge regions
Added edge caching in Frankfurt, São Paulo, and Singapore. Most lesson lookups now return in under one network round-trip.
Streamable HTTP replaces legacy SSE transport
Migration complete. All clients should update to the new endpoint. Old SSE endpoint deprecated but functional until June.
Pricing
Start free. Scale when you're ready.
Closed beta in progress — paid tiers activate at public launch. Join the Community tier to start now.
Community
For indie devs and small teams.
Get API keyCurated
For teams shipping with AI agents daily.
Join closed betaEnterprise
For regulated industries with data-residency needs.
Talk to salesStop teaching your AI
the same lessons twice.
A growing library of curated AI workflow knowledge — available to your agents the moment they need it.