ProvenDone
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@ProvenDoneback up my photos and verify the copies"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
DID IT — done AND proven done
Alexa+ track entry · Build, Ship, Shape: Amazon Developer Hackathon 2026
Ask an assistant to do something and it says "OK, done." But did it?
DID IT is a self-hosted MCP server for Alexa+ where every task ends in
a receipt: the plan, each step's execution, and machine-checkable
evidence — sha256 pairs, before/after listings, live HTTP responses.
An action is only done when it is proven done.
MCP spec
2025-11-25negotiated live on the wire (the Alexa+ track requires 2025-11-25 or later)Transport: Streamable HTTP (stateless) — the same path Alexa+ uses
7/7 unit tests + end-to-end smoke PASS (
pytest tests/,python scripts/smoke_e2e.py)
Quick start
./scripts/run_local.sh
# MCP server -> http://127.0.0.1:8765/mcp
# voice demo -> http://127.0.0.1:8770The demo UI has one-click tasks and a mic button (browser speech) — say "back up my photos" and watch the receipt build: every copied file with source/copy sha256 pairs compared in the same run.
Related MCP server: Orchestrator Python MCP Server
The tools
MCP tool | What it does |
| Run a multi-step personal-ops task → returns a receipt (plan, per-step evidence, verdict) |
| Re-read a stored receipt on demand — the proof, whenever you want it |
| Most recent receipts: id, task, verdict |
Verified executors (every one returns action, ok, summary, evidence, verified):
Executor | Proof it must produce |
| per-file sha256 of source and copy, byte counts, |
| HTTP status, latency, body sha256, text-found index |
| listing, count, total bytes (the evidence primitive) |
| before/after listings + per-move existence checks |
| disk usage + platform facts |
A receipt (from the passing smoke run)
{
"receipt_id": "rcpt_1769296393_a1b2c3d4",
"task": "smoke: back up the test suite",
"verdict": "done-and-proven",
"steps": [{"step": 1, "action": "file_backup", "verified": true,
"evidence": [{"file": "test_core.py", "bytes": 3496,
"sha256_source": "e98d…7225",
"sha256_copy": "e98d…7225", "match": true}]}],
"duration_ms": 2127
}Three verdicts, and no way to fake the top one: done-and-proven (every step verified), done-unverified (ran, proof missing), failed (honest, recorded, never hidden — unparseable tasks return a receipt that says so).
Architecture
Alexa+ ──┐
│ Streamable HTTP
Browser ─▶ demo/bridge.py ──▶ server/app.py (MCP server, FastMCP)
run_task ─▶ planner ─▶ verified executors
│ (Bedrock if AWS creds, else rules)
▼
receipts/*.json (the proof, on disk)server/app.py — FastMCP server, Streamable HTTP, stateless
server/executors.py — the verification discipline (hash/list/HTTP proofs)
server/planner.py — LocalPlanner (offline rules) + BedrockPlanner (AWS mini)
demo/ — voice-first web UI + tiny bridge that speaks real MCP
tests/ + scripts/smoke_e2e.py — the receipt pipeline is test-proven
Docs
docs/ALEXA_PLUS_SETUP.md — expose this server to Alexa+ (tunnel + config)
docs/AWS_BUILDER.md — Amazon Bedrock planner integration (mini challenge)
docs/DEMO_SCRIPT.md — the 3-minute demo storyboard
FRICTION_LOG.md — friction entries logged while building (10% bonus lane)
PRODUCT_FEEDBACK.md — per-tool feedback, updated on real use
Why this wins on the criteria
Tech Implementation — a real MCP spec
2025-11-25Streamable HTTP server (the track's literal requirement), evidence-bearing tools, tests that tamper data and expect the system to catch it.Design — voice-first interaction; the receipt is a first-class product object; failures are honest and explained.
Potential Impact — "did it actually do it?" is the #1 consumer-agent trust gap; receipts generalize to any executor you add.
Quality of the Idea — the assistant that won't claim success without proof. The LLM (Bedrock, optional) may propose; only the machine verifies.
MIT License — see LICENSE.
This server cannot be deployed
Maintenance
Related MCP Connectors
Human-in-the-loop approval for agent actions, with verifiable action-bound receipts.
Issue & verify signed (ed25519), hash-chained, timestamped provenance receipts for agent actions.
Independent effect verification and signed receipts for consequential AI agent actions.
Governed AI actions with signed, verifiable receipts: free keyless reads, human-approved writes.
Related MCP Servers
- AlicenseAqualityBmaintenanceEnables AI agents to create cryptographically verifiable receipts of their delegated work, with capabilities for multi-party approval and offline verification.1150 npmApache 2.0
- FlicenseBqualityBmaintenanceEnables AI coding assistants to run a machine-verified DESIGN→PLAN→EXECUTE→VERIFY→COMPLETE workflow with human approval gates, state integrity checks, and DAG task scheduling.7-
- AlicenseNot gradedqualityCmaintenanceEnables Alexa+ agents to maintain auditable operational continuity across shifts by turning speech into verifiable state, persisting unresolved work, refusing unverified actions, and requiring human approval before executing and confirming high-risk tasks.MIT
- AlicenseNot gradedqualityBmaintenanceEnables Alexa+ to collect recent security and IoT context, reason locally, prepare bounded physical actions for explicit approval, execute them through adapters, and return verified evidence in real time.MIT