Use configured inference model for MCP assists

This commit is contained in:
Codex
2026-04-29 16:03:13 +00:00
parent 50e8775276
commit 524fb29ed5
6 changed files with 78 additions and 26 deletions
+6 -5
View File
@@ -1,8 +1,7 @@
import { NextResponse } from "next/server"
import { resolveOpenAiCompatibleBaseUrl, resolvePrimaryInferenceModel } from "@/app/lib/inferenceModel"
const REQUEST_TIMEOUT_MS = Number.parseInt(process.env.MCP_REGISTRY_ASSIST_TIMEOUT_MS || "45000", 10)
const MODEL = process.env.MCP_PREFLIGHT_LLM_MODEL || process.env.OPENAI_MODEL || "gpt-4o-mini"
const BASE_URL = (process.env.MCP_PREFLIGHT_LLM_BASE_URL || process.env.OPENAI_BASE_URL || process.env.VLLM_BASE_URL || "http://localhost:8000/v1").replace(/\/+$/, "")
function extractJsonObject(value: string) {
const trimmed = value.trim()
@@ -19,8 +18,10 @@ export async function POST(request: Request) {
const body = await request.json().catch(() => ({}))
const prompt = typeof body?.prompt === "string" ? body.prompt.trim() : ""
if (!prompt) throw new Error("registry prompt is required")
const model = await resolvePrimaryInferenceModel()
if (!model) throw new Error("No primary inference model is configured")
const response = await fetch(`${BASE_URL}/chat/completions`, {
const response = await fetch(`${resolveOpenAiCompatibleBaseUrl()}/chat/completions`, {
method: "POST",
signal: controller.signal,
headers: {
@@ -30,7 +31,7 @@ export async function POST(request: Request) {
: {}),
},
body: JSON.stringify({
model: MODEL,
model,
temperature: 0.1,
response_format: { type: "json_object" },
messages: [
@@ -56,7 +57,7 @@ export async function POST(request: Request) {
baseUrl: typeof suggestion?.baseUrl === "string" ? suggestion.baseUrl : "",
description: typeof suggestion?.description === "string" ? suggestion.description : "",
},
model: MODEL,
model,
})
} catch (error) {
const message = error instanceof Error ? error.message : "Failed to assist registry setup"