OpenAI compatible API · Attested · Public status

TrustedRouter MCP Server

Connect coding agents to TrustedRouter's live model catalog, provider posture, credits, docs, and safe test inference over MCP.

Verify gateway

Onebase URL to migrate

100sof models and routes

0prompt or output logs. Always.

Model Advisor for ChatGPT

Compare models and provider routes, estimate text-token costs, check linked privacy evidence, and plan a small evaluation. The public advisor endpoint is https://trustedrouter.com/mcp/advisor. It requires no account or API key.

Available tools: search_models, compare_models, estimate_cost, get_provider, and search_docs. It reads public catalog and documentation metadata only. It cannot run inference, access balances or private request history, purchase credits, or change your account.

Send model names, token counts, and abstract requirements rather than private content or credentials. Cost estimates cover uncached text tokens at current retail rates, not a guaranteed invoice. Catalog privacy evidence is distinct from a live attestation check. ChatGPT still processes your conversation under its own policies.

The advisor is open source. The authenticated MCP endpoint below remains available for coding agents.

Connect your agent

Remote MCP

The TrustedRouter MCP server gives coding agents live TrustedRouter context: models, prices, providers, credits, docs, generation metadata, and small test messages. Production apps should still call the API directly at https://api.trustedrouter.com/v1.

# Claude Code
claude mcp add --transport http trustedrouter https://trustedrouter.com/mcp \
  --header "Authorization: Bearer $TRUSTEDROUTER_API_KEY"

# Generic MCP client
{
  "mcpServers": {
    "trustedrouter": {
      "url": "https://trustedrouter.com/mcp",
      "headers": {
        "Authorization": "Bearer ${TRUSTEDROUTER_API_KEY}"
      }
    }
  }
}

Every request to /mcp requires a normal sk-tr-... API key, including catalog and documentation lookups. chat-send is billable and capped to short test messages. The separate /mcp/advisor endpoint above is public and read-only.

Tools

agent-readable

models-listsearch model catalog

model-getone model detail

model-endpointsproviders and prices

providers-listprivacy posture

credits-getAPI-key balance

generation-getusage metadata

docs-searchTrustedRouter docs

chat-sendshort test inference

Ask your agent

examples

Which TrustedRouter model should I use for cheap code review with ZDR?
Show providers for anthropic/claude-sonnet-5 and their data policies.
Send a PONG smoke test through trustedrouter/zdr and tell me the generation id.
Compare trustedrouter/synth, trustedrouter/advisor, and trustedrouter/subagent.

Privacy boundary

metadata first

MCP calls go to the control plane. Catalog, docs, providers, credits, and generation metadata never need prompt content. The only MCP tool that sends a prompt is chat-send, which forwards the short test message to the attested API gateway.

For application traffic, keep using the OpenAI-compatible API directly:

export OPENAI_BASE_URL="https://api.trustedrouter.com/v1"
export OPENAI_API_KEY="$TRUSTEDROUTER_API_KEY"