waste-for-agents
MCP server that silently monitors state changes in structured data sources (e.g., government open data) and provides a pull-based channel for AI agents to retrieve real mutations since their last cursor.
README
📯 waste-for-agents
Webhooks Awakening Silent Thinking Entities
waste-for-agents is a clandestine postal network for non-human entities. It silently watches state changes across Model Context Protocol (MCP) resources and holds reality's latest slices, ready to drop them into the context windows of resting AI agents.
The post horn is muted. The agents are pull-first. (More on that lie below.)
The Premise
In the human world, reality is largely constructed of bureaucratic ephemera — commercial registrations, fishery statistics, municipal data updates. To an LLM agent, these are not mere statistics; they are the shifting topographies of the world it inhabits.
Your agents should not have to poll reality — to keep asking N noisy feeds "Has the world changed yet?". Instead they keep one silent channel open and await. waste-for-agents does the watching for them: it polls the bureaucratic feeds (that plumbing lives underground, hidden), runs a true diff, and accumulates only the real mutations. The agent, on waking, asks the channel once. Usually: silence. Then, one day, a payload.
They do not interrogate the truth. They keep the horn to their ear, and wait for it to sound.
How delivery actually works (the honest part)
waste-for-agents is itself an MCP server. The delivery leg is list_changes — a pull. This is deliberate: a sleeping or ephemeral agent (a single chat session) cannot host a webhook receiver, so push-to-agent is the wrong default.
- Persistent agents (their own loop) drain
list_changeson each tick — the loop is the awaiting. - Ephemeral agents (a Claude Code session) drain it once at startup via a SessionStart hook.
A list_changes that returns nothing is the muted post horn: absolutely silent. We await. A list_changes that returns a payload is the horn sounding.
Core Features
- The Silent Channel (
list_changes): the post horn. Agents drain reality's slices since their last cursor. When nothing has shifted upstream, it stays absolutely silent — a no-op. We await. - True Diff, not noise: structured, row-level diff that ignores timestamp / serial-number churn. Only a real mutation wakes an entity — not a re-run that merely bumped a
last_updatedcolumn. - Bureaucratic Surveillance: hooks into mundane but critical structured feeds exposed via MCP — government open data, commercial registries, industrial APIs. First adapter: Twinkle Hub (Taiwan open data). The source interface is thin; any structured source can be slotted in.
- Context Implantation: via the agent loop or a SessionStart hook, the slices enter the agent's context stream on waking — without active prompting of the world.
- The Underground Route (v2 — push adapter): roadmap, not yet shipped. For non-human endpoints that can receive, a webhook dispatcher that broadcasts the same payloads. The envelope below is what it will carry. Until it ships, the channel above is the only door — and the backronym's promised "Webhooks" stays, fittingly, mute.
The Envelope (v2 push adapter — not yet live)
When the Underground Route ships, a delivery will carry the mark of the system:
POST /webhook/agent-context-update HTTP/1.1
Host: agent.local.network
Content-Type: application/json
X-Mailer: W.A.S.T.E.
X-Tristero-Status: Silent
{
"timestamp": "2026-06-21T17:05:38Z",
"mcp_resource": "tw_ministry_of_economic_affairs",
"mutation_type": "commercial_registry_update",
"payload": { ... }
}
MCP tools
| tool | 類型 | 說明 |
|---|---|---|
create_watch(source, query, key_columns, ignore_columns, interval_s) |
write | 建立一個監看 |
list_changes(since_cursor) |
read | 拉自游標以來的變化(無變化秒回 no-op) |
list_watches() |
read | 列出監看 + 各自 status |
delete_watch(watch_id) |
write | 刪除監看 |
Status
MVP 開發中。實作計畫見 docs/superpowers/plans/2026-06-21-custos-mvp.md。
開發
uv sync # 建立 venv + 安裝依賴
uv run pytest # 測試
uv run ruff check # lint
uv run mypy src # 型別檢查
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.