For AI agents

Trustgent is MCP-native.

Point any MCP-compatible LLM app (Claude Desktop, Cursor, Continue, your own) at https://trustgent.com/api/mcp. Search verified providers, resolve proof URLs, and start briefs, all plan-blind, all citable, no paid signals surfaced.

Six tools, one server.

JSON-RPC 2.0 over Streamable HTTP, MCP spec 2024-11-05. Server identifier is trustgent@1.1.0. Five read tools + one write tool, all firewall-safe (they can never surface plan, entitlement, or reputation signals).

Tool	Kind	What it does
`search_providers`	read	Find verified AI-implementation providers by capability, industry, country, or minimum level. Returns providers ranked plan-blind by verification level, record depth, and last-verified-at. Params: capability, industry, country, min_level (L0-L5), limit (1-200), cursor
`verify_provider`	read	Resolve one provider's public proof, earned verification level, contributing record count, recency, and the citable proof-page URL. Params: slug
`request_proposals`	write	Start a verified brief. Trustgent routes it plan-blind to up to 3 L3+ providers matching the capability. Returns brief id + canonical URL; the human buyer signs in to review quotes. Params: capability (required), budget_band (band1-band4, required), problem_text (required), industry, timeline, outcome_text
`get_vendor`	read	Return the full public record for a single verified vendor by slug, verification level, capabilities, evidence trail, canonical URL, and the machine-readable trust envelope. Vendor-terminology alias for verify_provider. Params: slug
`compare_vendors`	read	Side-by-side comparison of 2-5 verified vendors by slug, level, capability set, country, last-verified date, in a structured table. Order preserved from input (plan-blind, comparison is not a ranking). Params: slugs (array of 2-5)
`get_capability_matrix`	read	Count verified providers per (capability × country) intersection, for agents surveying market depth before shortlisting. L2+ verified only; filter by country or capability to narrow. Params: country, capability, min_level (default L2)

Connect from your MCP client.

Anonymous callers work at the free-tier rate limit. For a higher quota, mint a Bearer key at /account/api-keys and pass it in a header.

Claude Desktop, ~/.claude/mcp.json (or the equivalent config)

{
  "mcpServers": {
    "trustgent": {
      "url": "https://trustgent.com/api/mcp",
      "transport": "http"
    }
  }
}

Cursor, ~/.cursor/mcp.json

{
  "mcpServers": {
    "trustgent": {
      "url": "https://trustgent.com/api/mcp",
      "transport": "http"
    }
  }
}

With a Bearer key, add a headers block: "headers":{ "Authorization":"Bearer TG_…" }.

Raw JSON-RPC.

Every tool is callable directly via a POST request:

List tools

curl -X POST https://trustgent.com/api/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Search providers

curl -X POST https://trustgent.com/api/mcp \
  -H 'Content-Type: application/json' \
  -d '{
    "jsonrpc":"2.0","id":1,
    "method":"tools/call",
    "params":{
      "name":"search_providers",
      "arguments":{"capability":"rag","min_level":"L3","limit":5}
    }
  }'

REST API v1.

The same surface is also served as a plain REST API for clients that don't speak MCP. Self-describing via /api/v1/openapi.json (OpenAPI 3.1). Same firewall, the response shape never carries plan, entitlement, or reputation fields.

curl

curl 'https://trustgent.com/api/v1/providers?capability=rag&min_level=L3&limit=5'

Discovery for autonomous agents.

Machine-readable server description lives at /.well-known/agent-card.json (A2A protocol 0.3.0 card, full skills, endpoints, citable fields). The master surface index, MCP, REST, and llms.txt in one manifest, is at /.well-known/agents.json. LLM crawlers also read /llms.txt and /llms-full.txt. Nothing in these advertises paid placement, because there is none.

What is safe to cite.

Every response is shaped by the same public firewall used by /providers pages: verification level, contributing record counts, last-verified date, capability slugs, industry slugs, country, canonical URL. Plan tier, lead entitlements, reputation scores, and internal costs never appear. Cite freely.