Rostra
Rostra is a portable Agent Plugin for the unified sovereign-search and research data
layer behind the Rostra MCP endpoint (https://ask.daft.onl/mcp): web search, page
reading, news, prediction markets, Reddit, Hacker News, Mastodon, Bluesky, X, GitHub,
and YouTube behind one typed result shape.
This plugin ships no server code and no local state. It is a declaration package:
one mcp.json remote-server entry plus eight Agent Skills that teach when and how to
use the Rostra tools. All capability lives upstream; the plugin makes it discoverable
and usable by agent clients.
Skills
| Skill | What it is for |
|---|---|
| topic-research | deep research: strands, primary sources, verification, synthesis |
| person-research | identity-first briefs on people, orgs, projects, products |
| investigation | trace a claim to its origin; grade what the receipts support |
| vibe-check | how different communities actually received something |
| repo-intelligence | OSS due diligence: maintained, adoptable, worth it? |
| trend-explainer | why something is blowing up, and whether it is real |
| video-digest | video to usable knowledge, currency-checked |
| find-anything-online | recover the exact thing from messy half-remembered clues |
Each skill declares allowed-tools scoping exactly the Rostra subset its job needs.
Authentication
The portable files contain no credentials (Agent Plugins v1.0.0: configured
headers must be literal, secret-free package data). Rostra's MCP endpoint is
OAuth-authenticated today; a shared API key path is being added upstream with an
explicit MCP opt-in per key. Clients choose their own auth (see DESIGN.md §4).
Sources of truth
- Agent Plugins spec + author guide: https://agent-plugins.org/specification, https://agent-plugins.org/plugin-authors
- Rostra tool contracts:
docs/MCP_TOOLS.md(upstream Rostra repo) - Design + auth model + milestones:
DESIGN.md