OPTINROUTER / DEVELOPER DOCS

MCP integration preview

A scoped agent interface to routing, policy explanations, synthetic action simulation and trace metadata. Production MCP access is closed.

Contract preview · Production API access is closed. All /v1 endpoints currently return 401.Install CLI
01

Transport and current availability

The proposed endpoint is https://optinrouter.ai/mcp. All methods currently return HTTP 401, including supplied keys. The downloadable manifest is a design artifact, not a running MCP server. This compatibility profile pins MCP 2025-11-25 Streamable HTTP and its initialization lifecycle; it does not claim support for every newer protocol version or every client. Origin validation, version negotiation, tenant authentication and protocol conformance must pass certification before access opens.

json
{
  "url": "https://optinrouter.ai/mcp",
  "profile": "2025-11-25-compatibility-preview",
  "status": "access_closed"
}
02

Tools and authority

Agents may request only approved tools with narrow projections. Tool annotations are descriptive; authorization is enforced by the server using tenant identity, required scopes and active policy. The initial tool set provides route explanation, synthetic graph simulation, policy validation, trace metadata aggregate usage, buyer offer eligibility and provider certification requirements. It excludes shell execution, arbitrary network requests, purchases, messages, policy activation and data delivery.

ToolScopeInput / output
routes_explainrouting:readPinned policy + candidate IDs → eligibility/reasons
actions_simulateactions:simulatePinned graph + named synthetic fixture → event metadata
policies_validatepolicies:validatePolicy document → validation issues
traces_gettraces:readTrace ID → authorized metadata
usage_listusage:readBounded filters → aggregate usage
buyer_offer_previewmarketplace:readDataset + purpose + region → consent/license eligibility; no purchase
provider_certification_previewprovider:validateEndpoint + profile → capability checks; no registration
03

Context and action isolation

A request or model response is untrusted data. It cannot register tools, grant scopes, change location policy, expand a projection or authorize effects. The action graph pins tool revisions, dependencies, phases, output slots, concurrency and branch deadlines. Required unknown or failed controls block release. Synthetic action simulation performs no provider inference and no external side effects; production execution is a separate gateway operation with its own usage accounting.

04

Errors, cancellation and evidence

The future authenticated MCP layer must map policy denial, unsupported grammar, stale revision, missing regional evidence, scope denial and cancellation into structured bounded results. The compatibility transport uses JSON-RPC, version headers and explicit cancellation notifications. Closing a connection alone does not establish cancellation. Logs and resources exclude raw content by default. Conformance cases cover cross-tenant IDs, prompt-injection attempts, output-slot conflicts, cancellation after dispatch and absent evidence.

json
{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "routes_explain",
    "arguments": {
      "policy_ref": "pol_engineering@3",
      "candidate_ids": [
        "eu-fra"
      ]
    }
  }
}

Contracts & examples

Download the preview, validate its shape locally and review the capability limits before planning an integration.

OpenAPI preview

OptinRouter AI guide

Product and developer guidance.

A good place to start

What would you
like to know?

Ask about public APIs, policies, grammars, context actions or the product. This guide uses approved information and can’t reveal private details or take actions.

Recent messages stay in this page’s memory. Reload or reset to clear them here.