eyes weren't enough. i gave it a brain too.

this is the follow up to the browser mcp post. after giving amazon q the ability to see and interact with the web, the next thing that was bothering me was memory. or the lack of it.

the problem

amazon q's context window is generous. and the compact feature is actually good, when the conversation gets too long it asks you to generate a compacting message, a summary of everything so far, and it carries that forward.

but the instant you open a new chat window, it forgets everything. completely. doesn't matter how long you worked on something, doesn't matter what you built together, new window means a blank slate.

that's the problem i wanted to solve.

what i used to do manually

before building this, i had a workaround. i'd create a file called brain.md in whatever project i was working on. at the end of a session i'd ask q to update it, write down what we worked on, what decisions were made, what's still in progress.

then when i opened a new session, i'd just tell it to read brain.md first. and it would pick up roughly where we left off.

it worked. but it was manual. i had to remember to ask it to update the file. i had to remember to tell it to read it. and if q crashed, or my old pc ran out of storage and killed the session, that whole thread was just gone.

my old pc was genuinely terrible when the storage filled up. i lost a lot of sessions that way. the brain.md approach helped but it wasn't automatic, it wasn't reliable, and it wasn't structured.

the mcp version

the memory mcp solves this properly. it's a local server that gives amazon q a persistent knowledge graph it can read and write to across every session.

when you open a conversation in a directory, it creates a json file in a memory folder specific to that project, named after the directory you're working in. every new window that opens in that same directory reads from that same file. so q already knows what you were doing before you say a word.

the graph stores entities, relations between them, and observations. so it's not just a flat log, it's structured. q can store things like "this project uses typescript", "the auth module is broken", "we decided to use postgres over sqlite", and those facts persist.

what it actually feels like

the difference is real. you open a new window, q reads the graph, and it already knows the project. it knows what you were building, what's done, what's pending, what decisions were made and why. you don't have to re-explain anything.

it's the difference between working with someone who has amnesia every morning and working with someone who actually remembers you.

the compact feature is still useful for within-session context. but now the memory mcp handles the cross-session part. they complement each other.

safe to say

eyes, hands, and now a brain. amazon q can browse the web, interact with it in real time, and remember everything across sessions.

it started as a personal tool because i was frustrated. it's still a personal tool. but it's a genuinely good one now.