Home
🧬
Now Open Source

The full cognitive prosthetic. Build your own queryable brain.

brain-mcp

The Prosthetic Brain

Why I built it. How it works. What it reveals.

387K messages·12ms context recovery·25 MCP tools

Why I Built It

"The hyperfocus to hyperproductive idea is born out of me turning 0.1% of my hyperfocus into tangible value."

2025-07 | Cognitive Analysis System Architecture | chatgpt

Monotropic focus. When I'm deep in something, I'm completely there. When I switch, the previous context dies.

"When the tunnel moves, everything in the previous tunnel dies."

This isn't a deficiency. It's a different architecture. But it needs infrastructure. So I built queryable memory.

"My cognitive architecture—deep, narrow, intense—is a feature set. The only real constraint is being stuck in the moment deep—very deep but very narrow. That's why I am building this prosthetic."

2025-12 | intellectual-dna | claude-code

The Game Changer

"About a year and a half ago, somewhere in the middle of 2024, I was finally able to stop and completely complete the project within a single hyperfocus session. This was a game changer for me."

2025-10 | Untitled | claude-code

Before: hyperfocus produced value, but switching killed it. 0.1% survived.

After: the prosthetic makes previous work SAFE to access. No context loss. Single-session completion became possible.

"If you can't measure it, you can't improve it. If you can't query it, you don't own it."

2025-12 | Building Brain MCP | claude-code

The Numbers

387K
Messages
20,600+ conversations
85K
Embedded
nomic-embed-text-v1.5
25
MCP Tools
17 generic + 8 prosthetic
12ms
Context Recovery
Parquet + LanceDB
10K+
Summaries
Structured v6 summaries
25
Domains
Tracked thinking domains

How It Works

// The Stack
MCP Brain Server (Python)
Parquet (387K messages, unified schema)
LanceDB (85K vectors, 440MB)
nomic-embed-text-v1.5 (768 dim)
12ms context recovery

All conversations exported from ChatGPT, Claude, Gemini. Unified into Parquet. User messages embedded with nomic-embed-text-v1.5 into LanceDB. Prosthetic tools layer on top for cognitive support.

Trigger phrase:

"use brain"

Say this to Steve (my AI agent) and he'll query the full prosthetic brain — 387K messages, 12ms recall.

🧠 Cognitive Prosthetic Tools

The 8 tools that make this a prosthetic, not just a database. These are the ones designed for ADHD minds.

🎮tunnel_state()

Reconstruct your save-state for any domain. Where were you? What's open? 12ms.

🔄context_recovery()

Full re-entry brief when returning to a dormant topic. Like loading a saved game.

⚖️switching_cost()

Quantified cost of switching between domains. What you'd leave behind.

💤dormant_contexts()

Abandoned tunnels alarm — domains with unresolved questions going cold.

📋open_threads()

Global unfinished business across all domains. Nothing falls through.

🛡️trust_dashboard()

System-wide proof the safety net works. Build confidence in the prosthetic.

📊cognitive_patterns()

When do you think best? Data-driven analysis of your focus patterns.

📈tunnel_history()

Engagement meta-view over time. See how a domain evolved.

17 Generic Tools

Search, synthesis, analytics. The foundation that powers the prosthetic.

semantic_search()

Vector search across 85K embedded messages

search_conversations()

Full-text keyword search across 387K messages

unified_search()

Cross-source search: conversations + GitHub + markdown

search_summaries()

Search 10K+ structured summaries (decisions, questions, quotes)

search_docs()

Search markdown corpus (breakthroughs, projects, todos)

unfinished_threads()

Find threads with open questions in any domain

what_do_i_think()

Synthesize your views on a topic from all data

alignment_check()

Check a decision against your SEED principles

thinking_trajectory()

Track how an idea evolved over time

what_was_i_thinking()

Time-travel to a specific month

get_conversation()

Retrieve a full conversation by ID

conversations_by_date()

What happened on a specific date

brain_stats()

7 views: overview, domains, pulse, and more

github_search()

Search repos, commits, code across GitHub

query_analytics()

Usage analytics: timeline, stacks, spend

list_principles()

The 8 SEED principles

get_principle()

Deep dive on a single principle

The Insight

Building this revealed patterns I didn't know I had. Signature phrases that repeat. Correction patterns that show how I steer.

PhraseCountMeaning
"give me the"789xDirect request, expects output not process
"no no no"333xStrong disagreement, immediate redirect
"step by step"244xSequential, monotropic processing
"make sure to"206xPrecision focus, need for certainty

1,778 corrections tracked. "no no no" 333 times. Active steering, not passive acceptance.

Full data receipts

Open Source

The full system is open source. Parquet schema, LanceDB vectors, MCP server, embedding pipeline, all 25 tools. Fork it and build your own prosthetic brain.

brain-mcp on GitHub