Explore capabilities →

MCP API

Every capability, from your own agent.

A public Model Context Protocol endpoint over the same connected records behind this site: 195 experts across 43 courts, the counsel on deals and offerings, federal buying, rules in progress, regulated sites and the companies that tie them together, in one knowledge graph. Read-only, no account, no API key.

POST https://uncommon-context.vercel.app/mcp

Connect

Point any MCP client at one URL.

Add it to a client's server list. There is no token to request and no sign-up step, because everything served is public record.

{
  "mcpServers": {
    "expert-edge": {
      "type": "http",
      "url": "https://uncommon-context.vercel.app/mcp"
    }
  }
}

Works with clients speaking MCP 2026-07-28 (stateless, per-request metadata) and with the earlier initialize-handshake revisions back to 2025-03-26. Call server/discover to read the supported list from the server itself.

Tools

13 read-only tools.

Nothing here writes, harvests, or reaches another host. Every tool is annotated read-only so a client never gates a lookup behind a confirmation it does not need.

search_experts

Search court-verified expert witnesses by name, field, institution, topic, court, or case caption. Every expert returned was named in a published court opinion. Use `list_filters` first to get valid industry and court values.

  • query — Free text: expert name, institution, research topic, court, or case caption.
  • industry — Industry code from list_filters, e.g. medical, engineering, finance.
  • court — Full or partial court name, e.g. 'Ninth Circuit'.
  • ruling_status — Filter by how courts ruled on the testimony.
  • min_cases — Only experts with at least this many court appearances.
  • limit — Max results (default 20).
  • offset — Skip this many results for paging.

get_expert

Full dossier for one expert: every opinion that named them with the naming sentence, Rule 702 grounds and outcomes, numbered exhibits with quotes, a cross-examination outline, scholarship, identifiers, industry payments, and any field withheld for weak identity evidence.

  • name — Expert name, exact or partial.
  • slug — Expert slug, as returned by search_experts.

list_filters

Every industry, court, and ruling-status value present in the data, with expert counts. Call this before filtering so you use values that actually exist.

  • query — Free text: an issue, statute, agency, office or firm, e.g. 'former EPA RCRA enforcement'.
  • include_near_misses — Include near misses with the criterion each missed (default true).
  • firm — Only lawyers at this firm, e.g. 'Latham'.
  • limit — Max lawyers (default 25).

database_stats

Totals (experts, appearances, distinct courts, year range) and the inclusion rule. Use it to state coverage accurately rather than guessing how complete the database is.

  • query — Free text: an issue, statute, agency, office or firm, e.g. 'former EPA RCRA enforcement'.
  • include_near_misses — Include near misses with the criterion each missed (default true).
  • firm — Only lawyers at this firm, e.g. 'Latham'.
  • limit — Max lawyers (default 25).

search_lawyers

Search a Law Firm Intelligence research snapshot for lawyers whose published evidence fits a need: government service, litigation, counseling, practices, admissions and publications. Returns who met the report's screen and who was a near miss, the passage that matched each term, and the firm-profile source. Also searches public biographies from the largest US firms: practices, offices, education, bar admissions, clerkships, languages and prior government service, each from the bio itself. A bounded index, not a full directory.

  • query — Free text: an issue, statute, agency, office or firm, e.g. 'former EPA RCRA enforcement'.
  • include_near_misses — Include near misses with the criterion each missed (default true).
  • firm — Only lawyers at this firm, e.g. 'Latham'.
  • limit — Max lawyers (default 25).

get_lawyer

Full record for one lawyer in the research snapshot: every relationship the report supports, what makes the candidate distinctive, the report's own evidence limitations, and the cited firm profile.

No parameters.

search_awards

Federal contract awards from USAspending.gov on the issues the suite tracks (PFAS, data centers, environmental remediation, air permitting): who was paid, by which agency, for what, and when. Every result links to its USAspending record.

  • query — Free text over recipient, description and agency, e.g. 'PFAS Air Force'.
  • agency — Awarding agency or sub-agency, e.g. 'Environmental Protection Agency'.
  • min_amount — Only awards of at least this many dollars.
  • limit — Max results (default 10).

search_federal_register

Federal Register rules, proposed rules and notices on the tracked issues (PFAS, CERCLA, stationary engines such as data-center backup generators, data centers, environmental impact statements in Virginia), with dockets, comment deadlines and the federalregister.gov link.

  • query — Free text over title, abstract, agency and docket.
  • agency — Agency name, e.g. 'Environmental Protection Agency'.
  • type — Document type.
  • limit — Max results (default 10).

search_facilities

Facilities in EPA's ECHO database for the tracked places (today: Virginia data centers), with address, county, the environmental programs that regulate them, inspection and compliance fields, and the ECHO facility report link.

  • query — Free text over name, address, county and programs.
  • county — County, e.g. 'Loudoun'.
  • program — Program, e.g. 'Clean Air Act'.
  • limit — Max results (default 10).

explore_graph

Look up any entity in the Uncommon Context knowledge graph (a lawyer, expert, firm, agency, issue, place, award, rule or facility) and get its connections across every capability, each with the source sentence and link that supports it.

  • entity — Entity name or id, e.g. 'PFAS', 'Stacey Geis', 'agency:epa'.
  • capability — Only connections from this capability.
  • limit — Max connections (default 25).

company_dossier

Everything the suite holds on a public company: its profile, the counsel named on its deals and offerings, what it disclosed, its litigation and environmental reserves, recalls, lobbying, agency documents naming it, federal contracts, regulated sites, the expert cases that name it, the rules in play, its peers and the firms its peers use. Each part links to its record.

  • company — Company name, ticker or SEC CIK, e.g. 'Caesars' or 'CZR'.
  • limit — Max companies (default 3).

firm_dossier

A law firm as the public record shows it: the matters in which companies' own filings name it as counsel (role, practice, the sentence and the link), its clients with what each has disclosed or had published about it in the last year, the firms it worked alongside, the lobbying it files for tracked companies and its lawyers in the research. With no firm, ranks the firms the filings name most often, overall or in a practice.

  • firm — Firm name, e.g. 'Latham' or 'Kirkland & Ellis'. Omit to rank firms.
  • practice — With no firm: rank firms in this practice, e.g. 'restructuring'.
  • limit — Max results.

trace_connection

Shortest evidenced path from one entity to another entity, or to the nearest record of another capability. Each hop carries its source sentence. This is how context is carried from one capability to the next.

  • from — Start entity name or id.
  • to — End entity name or id.
  • to_capability — Or: the nearest record of this capability.

Try it

One request, no credentials.

curl -s -X POST https://uncommon-context.vercel.app/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -H 'MCP-Protocol-Version: 2026-07-28' \
  -H 'Mcp-Method: tools/call' \
  -H 'Mcp-Name: search_experts' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
       "params":{"name":"search_experts",
         "arguments":{"query":"toxicologist","limit":3},
         "_meta":{"io.modelcontextprotocol/protocolVersion":"2026-07-28"}}}'

Opening https://uncommon-context.vercel.app/mcp in a browser is expected to show a notice rather than data: a browser sends GET, and the endpoint answers POST. That is not an outage. Older clients that cannot reach a remote HTTP server directly can bridge with npx -y mcp-remote https://uncommon-context.vercel.app/mcp.

Reading the results honestly

What the fields do and don’t assert.

Every result carries source_status: when its records were refreshed and any part that failed, so “could not check” is never read as “none”. Matches made on a name say so. For expert records specifically:

An affiliation is a publication affiliation

affiliation_basis: "publications" means the institution printed on that expert’s papers. It is not a claim about current employment.

“No ruling” is not approval

ruling_status: "no_ruling" means an opinion named the expert but the court never ruled on the testimony.

Withheld is not absent

withheld lists fields suppressed because the identity match rested on a name alone. Check homonym_count before asserting identity.

An expert missing from the database means no downloaded opinion named that person in an expert role. That is unresolved, not a finding that they are not an expert. The full rules are served as the expertedge://ground-rules resource.