| name | searxng-kg-ingestion |
| skill_type | skill |
| description | Persist SearXNG search results into the epistemic-graph knowledge graph via the searxng-mcp MCP server. Use when the agent must turn a web search into durable, semantically-searchable KG memory — each result becomes a :Document (plus :SearchQuery and :SearchEngine typed nodes with :resultOf / :fromEngine links) so later reasoning can recall and cite it. Do NOT use for one-off reads you don't need to remember (use searxng-web-search) or to ingest a specific page's full text (use web-crawler). |
| license | MIT |
| tags | ["searxng","ingestion","knowledge-graph","documents","mcp"] |
| metadata | {"author":"Genius","version":"0.1.0"} |
SearXNG KG Ingestion
Native "maximum ingestion" of SearXNG results into the ONE epistemic-graph knowledge
graph. Runs a metasearch and pushes each result as a shared :Document (carrying
the result text + source_uri) alongside a :SearchQuery node and per-engine
:SearchEngine nodes, wired with :resultOf / :fromEngine. This builds a
recallable, citable corpus from the open web.
When to use
- Build durable KG memory of what the web says about a topic (for later recall/citation).
- Seed a research corpus that hub-side enrichment can chunk and embed for semantic search.
- Capture a monitoring snapshot you want to query again later, not just read once.
When NOT to use
- Throwaway lookups you don't need to persist →
searxng-web-search.
- Ingesting the full body of one known page →
web-crawler / web-fetch.
- Structured records from an enterprise system → that system's own ingest tool.
Prerequisites & environment
Connect via the mcp-client skill against the searxng-mcp MCP server. A reachable
epistemic-graph engine is required for anything to land; with no engine the tool is a
clean no-op (ingested: null). Same env matrix as searxng-web-search. Note:
SEARXNG_KG_INGEST (default-on) already ingests on every web_search — use the explicit
tool below when you want the ingest confirmation payload or ingest without a prior read.
Tools & actions
| Tool | Purpose |
|---|
searxng_ingest_search | Run a search AND push results to the KG; returns {listed, ingested}. |
web_search | Plain search; also ingests by default unless SEARXNG_KG_INGEST=false. |
Key parameters (searxng_ingest_search)
query — the search string (required); also becomes the :SearchQuery node.categories / engines / language / pageno — same semantics as web_search.
Recipes
Search + ingest a topic:
{"query": "zero trust network access self-hosted", "categories": ["it"]}
Ingest news coverage for a monitoring corpus:
{"query": "cilium 1.16 release notes", "categories": ["news"]}
Gotchas
- Best-effort by design:
ingested is null when no engine is reachable and the search
results are still returned — never assume persistence succeeded without checking it.
- Result
:Document ids are searxng:result:<url>, so re-ingesting the same URL MERGEs
(updates) rather than duplicating; the :SearchQuery id is a hash of query+language.
- Only results with usable text (title/snippet) are ingested; empty-snippet results are
skipped. The count in
listed may exceed the nodes actually written.
web_search already ingests by default — don't double-invoke both tools for the same
query unless you specifically want the {listed, ingested} confirmation.
Related
searxng-web-search / searxng-news-research — the read-side skills.kg-ingest (universal) — the broader knowledge-graph ingestion workflow.- The
searxng_mcp.ontology searxng.ttl defines the classes/links written here.
