11 tools that give an agent the whole web
One config block. Search, document retrieval, coverage transparency, and six graph operations that let an agent traverse relationships instead of rebuilding them in its context window.
{
"mcpServers": {
"unlob": {
"command": "npx",
"args": ["-y", "@unlob/mcp"],
"env": { "UNLOB_API_KEY": "ulb_your_key_here" }
}
}
}Coverage graph tools
Each of these collapses a pattern that would otherwise cost many searches and a great deal of in-context reasoning.
relatedReturns the connected neighbourhood of a passage from the coverage graph — the edges are what to read next.
Collapses: A search per hop. One call replaces the whole chain.
corroborateReports how many distinct hosts independently carry a story — the anti-hallucination check.
Collapses: Manual cross-searching to verify a claim.
authoritiesReturns the top passages on a topic ranked by graph centrality — trust-triage a field without reading junk into context.
Collapses: Reading twenty mediocre results to find the three that mattered.
dossierBuilds a one-hop brief on an entity: mentions, the hosts covering it, and the entities co-mentioned with it.
Collapses: ~10 searches plus a manual merge, in one call.
pathFinds the shortest chain of edges linking two passages — reasoning-path retrieval, or connect-the-dots.
Collapses: Open-ended investigation, reduced to a structural query.
assemble_contextGraphRAG as a service: returns a ready-to-read context pack — corroborated, story-deduplicated, trust-ranked and packed to a token budget, each passage carrying the reason it was included.
Collapses: The entire do-it-yourself RAG loop.
Search and retrieval tools
web_searchSearches the agent-first web index and returns metadata-only hits — url, host, snippet, score — never the page body.
get_documentFetches the full cleaned text of one passage by id — the lazy read path.
similarMore-like-this: given a passage id, returns its semantic neighbours.
browseBrowses without a query: the most recent, or highest-ranked, passages — optionally within one vertical.
why_notExplains why a URL is or is not in the index: present, removed with a reason, or never admitted.
Setup by client
The config is identical everywhere; only the file location differs.
Tell the agent about the graph tools
Models default to plain search and will not discover the graph tools on their own. A short instruction changes behaviour substantially:
Usecorroboratebefore stating anything as fact. Usewhy_notbefore concluding something does not exist. Useauthoritieswhen entering an unfamiliar topic. Useassemble_contextwhen a question needs many sources rather than one.
Frequently asked questions
What is the Model Context Protocol?
An open protocol for exposing tools and data to AI models through a uniform interface. One server implementation works with any compatible client — Claude Code, Claude Desktop, Cursor and others — instead of a bespoke integration per model and framework.
How many tools does the unlob MCP server expose?
11. Five for retrieval — web_search, get_document, similar, browse and why_not — and six for the coverage graph: related, corroborate, authorities, dossier, path and assemble_context.
Do MCP calls cost more than HTTP calls?
No. Every tool call is one API request billed against your quota, exactly as an HTTP call would be.
Why is my agent not using the graph tools?
Models default to plain search. A one-line system-prompt instruction naming corroborate, authorities and assemble_context and when to use them is usually enough to change that.
Where do I put the API key?
In an environment variable referenced from the config, not in the config file itself — project-scoped MCP configs are often committed. Keys are shown once and stored as hashes; if one leaks, rotate it.
Get a key and paste the config
10,000 requests a month, no card. The config above works unchanged once you have a key.