H herrington.ai
KnowMe Protocol · MCP 2024-11-05 · Voice-Seeded Human Context

LinkedIn is for humans to learn about humans.
KnowMe is for agents to learn about humans.

An agent can scrape your LinkedIn, but it's dry corporate boilerplate. It can scan your X feed, but it's fragmented noise. KnowMe MCP is your living machine-callable front door—kept fresh by an AI Interviewer that regularly interviews you on the unwritten layer and resolves questions queued by visiting agents.

Launch AI Interviewer Studio Read CUJ Case Study Inspect Live /mcp Endpoint

The Problem: Static Profiles Starve Autonomous Agents

When someone's agent reaches out to collaborate, recruit, or research you, here is what it sees today versus what KnowMe serves over MCP.

❌ What LinkedIn & X Give an Agent

Dry Resume & Fragmented Posts

Blocked by auth walls, stripped of personality, and frozen in time. Agents get job titles and generic skills, with zero insight into how you actually build, think, or collaborate.

{ "source": "linkedin_scrape_fallback", "name": "Daniel Herrington", "headline": "Chief AI Officer at The Zebra | Ex-Google", "skills": ["Artificial Intelligence", "Product Management", "Strategy"], "actual_daily_tools": null, "contrarian_beliefs": null, "how_to_pitch_or_collaborate": null, "can_agent_ask_followup": false }
✅ What KnowMe MCP Gives an Agent

Voice-Interviewed Living Context + Question Queue

Seeded by 5-minute Wispr Flow voice interviews. External agents query your verified stack, contrarian takes, and rituals—or call request_human_answer when they hit a gap.

{ "protocol": "knowme-mcp/1.0", "tool": "get_knowme_profile", "agent_workflow": "Google Antigravity IDE + Grok Bot over Tailscale Mac Mini mesh", "voice_dictation": "Wispr Flow (streams thoughts directly into specs)", "ai_hot_take": "Software's 'Napster moment' via open-source; voice input > voice output", "personal_texture": "Wrenches a Chevy Camaro in the garage; drinks Yerba Mate; 52\" monitor", "missed_call_tool": "request_human_answer(question, agent_name, context)" }

KnowMe Live Studio · herrington.ai/mcp

Every interview question is strictly classified by provenance: 🟢 Real Telemetry (Firestore logs), 🔍 Corpus Gaps (facts.json audit), or 🌱 Seeded by Us.

Question Provenance Audit: 🟢 Real Telemetry (Firestore Logs) 🔍 Corpus Gap (facts.json Diff) 🌱 Seeded by Us (Curated Prompts)
Scanner: scripts/run_knowme_gap_detector.py
🟢 Real Telemetry (Firestore) Source: Muse (Linux x86_64)
target: muse_assistant_role

What is 'Muse', how does it run on your Linux box, and how is its role different from Antigravity or Grok Bot in your personal agent stack?

Why this matters: Real Firestore telemetry shows two separate queries about 'Muse', yet 'Muse' is completely absent from facts.json.

✅ Compiled into Machine-Callable MCP Fact
🎙️ AI Interviewer Generated Follow-Up Question

1. Simulate an External Agent Calling /mcp

Inspect what Claude Code, Cursor, or an executive scout agent receives when querying Daniel's KnowMe front door.

🟢 Live Agent Missed-Call Queue

2. Queue a Real Question for Daniel's Next Interview

When an agent calls request_human_answer (or you submit below), it is logged with real_telemetry provenance in Firestore knowme_question_queue.

Live MCP JSON-RPC 2.0 Payload Inspector POST https://herrington.ai/mcp 

                

            

        


        
        

            

                

                    Human Context Coverage Radar (Computed from facts.json)
                

                

                    Tracks seeded keys vs. missing gap keys across 6 dimensions. Any missing key is automatically flagged by scripts/run_knowme_gap_detector.py.
                

            


            

                
            


            

                Completed Interviews & Provenance Log
            

            

                
            


            

                Verified Facts Currently Live on herrington.ai/mcp (0 keys)
            

            

                
            

        

    


    
    

        

            How the KnowMe Loop Works
        

        

            Instead of filling out static forms, KnowMe treats your personal context like an evolving codebase maintained through brief, high-leverage voice interviews.
        


        

            

                
1

                
Ingest the Public Baseline

                

                    Syncs your dry baseline from LinkedIn, GitHub repositories, X posts, and long-form essays so the interviewer never wastes your time asking what's already on your resume.
                

            

            

                
2

                
Audit Telemetry & Corpus Gaps

                

                    Scans real Firestore shell_conversations & knowme_question_queue logs alongside facts.json schema diffs—tagging every prompt with verifiable provenance.
                

            

            

                
3

                
5-Minute Wispr Flow Interview

                

                    On a regular cadence, the AI Interviewer pitches you 3–5 high-signal questions. You dictate stream-of-consciousness answers; Gemini compiles them into clean third-person facts.
                

            

            

                
4

                
Serve via Stateless MCP

                

                    Your updated human context graph is immediately queryable over Streamable HTTP (POST /mcp) by Claude, Cursor, Antigravity, and peer agents worldwide.