thorbit-knowledge-graph-mcp

MCP server for the Thorbit Knowledge Graph entity-linking pipeline. Hosted Thorbit crawls a site, extracts entities via TextRazor, and links them to Wikidata/Wikipedia/DBpedia/Freebase into a canonical entity library — this package is a thin stdio client that forwards every tool call to the hosted Thorbit route and is authenticated + metered against your Thorbit MCP API key.

This is a new package name. It does not replace or collide with the older, separate thorbit-knowledge-graph npm package (a standalone local-engine CLI) — installs of that package are unaffected.

Install / run

npx -y thorbit-knowledge-graph-mcp@latest

Or add it to an MCP client config:

{
  "mcpServers": {
    "thorbit-knowledge-graph": {
      "command": "npx",
      "args": ["-y", "thorbit-knowledge-graph-mcp@latest"],
      "env": {
        "THORBIT_KG_API_KEY": "thbt_mcp_..."
      }
    }
  }
}

Generate the same JSON from the CLI:

npx -y -p thorbit-knowledge-graph-mcp@latest thorbit-knowledge-graph-mcp-install --json --api-key thbt_mcp_...

Prefer a key file on shared machines and servers:

mkdir -p ~/.config/thorbit
printf '%s\n' 'thbt_mcp_...' > ~/.config/thorbit/knowledge-graph-mcp-key
chmod 600 ~/.config/thorbit/knowledge-graph-mcp-key
npx -y -p thorbit-knowledge-graph-mcp@latest thorbit-knowledge-graph-mcp-install --key-path ~/.config/thorbit/knowledge-graph-mcp-key

Run npx -y -p thorbit-knowledge-graph-mcp@latest thorbit-knowledge-graph-mcp-install --help for all options.

Environment

• THORBIT_KG_API_KEY — your Thorbit MCP API key (required). Fallbacks: THORBIT_API_KEY, THORBIT_MCP_API_KEY, or a key file at ~/.thorbit-knowledge-graph-mcp-key (override path with THORBIT_KG_MCP_KEY_PATH).
• THORBIT_KG_BASE_URL — optional, defaults to https://thorbit.ai. Fallback: THORBIT_BASE_URL.

The key is never logged.

Tools

1. kg_build_library — turn web page content into a linked entity library: extracts every entity and links it to Wikidata/Wikipedia/DBpedia/Freebase, dedupes, cleans, and ranks. Feed it scraped pages, or a url/urls to self-fetch (up to 500 URLs). Starts a durable run; returns { runPublicId, poll }.
2. kg_emit_schema — generate finished schema.org JSON-LD for one page (home/service/about/blog) from a page's content plus a built or saved library. Starts a durable run; returns { runPublicId, poll }.
3. kg_emit_schema_bulk — the bulk version of kg_emit_schema: emits JSON-LD for many pages in one call, in parallel, against a shared business + library.
4. kg_resolve_term — resolve a single term or phrase to its canonical knowledge-graph IDs (Wikidata QID, Wikipedia, DBpedia, Freebase mid) with no crawl. Synchronous.
5. kg_get — poll a kg_build_library / kg_emit_schema / kg_emit_schema_bulk run for lean status, phase/progress, and the artifact manifest.
6. kg_library_list — list the saved KG catalog (approved only by default; pass includePending to see the review queue).
7. kg_library_get — fetch one saved KG's full entity list by name, for review.
8. kg_library_save — save a built library into the catalog under a name, as PENDING (not usable for schema generation until approved).
9. kg_library_approve — approve (or un-approve) a saved KG so it can be referenced by name from kg_emit_schema/kg_emit_schema_bulk. A human decision — never call this on your own initiative.
10. kg_library_remove — remove a saved KG from the catalog.

It also exposes a resource template thorbit-kg://runs/{runPublicId} (run status as JSON) and a kg_intake prompt describing the build→poll contract and pricing.

Async build → poll flow

1. Call kg_build_library (or kg_emit_schema / kg_emit_schema_bulk). It returns quickly with a runPublicId and a poll target. Do not block.
2. Poll kg_get with that runPublicId until status is completed or failed.
3. When complete, the run's result/manifest is available via kg_get's response — read the library or generated schema.org JSON-LD from there.

Pricing is flat and charged upfront per tool call: 1,000 credits per kg_build_library run, 500 credits per page emitted via kg_emit_schema/kg_emit_schema_bulk. A failed run is refunded in full.

Develop

npm install
npm run typecheck
npm run test
npm run build