Skip to main content
The Solya MCP gateway lets an AI agent (Claude, Cursor, or any MCP client) drive Solya through the REST API — discover endpoints, inspect them, and call them — all behind three generic tools.

How it works

The gateway is a hosted service in front of the Solya API. It reads the live OpenAPI spec at runtime (so it never drifts from the API) and exposes every operation through three tools:
ToolWhat it does
search_operationsKeyword-search the API’s operations; returns a compact list (operation_id, method, path, summary, tags). Results are filtered to what your token can access.
describe_operationReturn the full schema of one operation — method, path, parameters, request/response bodies, required auth.
invoke_operationExecute an operation — fills path placeholders, adds query params and a JSON body, attaches your token, and returns the HTTP status + parsed response.
A typical agent flow:
1

Search

search_operations("rebalance plans") → find the relevant endpoint.
2

Describe

describe_operation("listRebalancePlans") → get its exact schema.
3

Invoke

invoke_operation("listRebalancePlans", query: { page: 1 }) → fetch the data.

Authentication & scope

The gateway is a thin proxy — it stores no secrets. Each client sends its own service-account token (solya_sa_…) as a bearer header, which the gateway forwards to the API. Therefore:
  • Everything runs as your token: same organization scope, same permissions. The agent can only do what the token’s permissions allow (the API still enforces 403).
  • search_operations hides operations your token can’t use (ergonomic, not a security boundary — the API is the boundary).
  • Responses are the standard API envelope: { "success": true, "data": … } or { "success": false, "errorCode": … } (see Error codes).
Because the gateway reflects the OpenAPI spec, any Solya endpoint your token can call is reachable through invoke_operation — no per-tool wiring. To connect a client, see Connect a client.