Integrations

Connect PromptFlowEngine

The same engine runs in every integration. Choose by where your prompts are: in a browser or editor (the extensions), inside an AI assistant or agent (MCP), or in your own code (HTTP API). None of them calls a language model, and none needs an account or API key.

Option 1: MCP server

PromptFlowEngine ships an MCP server (v2.1.0); your AI tool is the client. The transport isstdio only: the client starts node packages/mcp-server/dist/bin.js on your machine. There is no hosted MCP endpoint and no authentication, because only a process you start can talk to it.

Prerequisites

• Node.js 22 or later (Windows, macOS, or Linux) with Corepack.
• A checkout of the PromptFlowEngine repository (npm publishing is pending). Build it once:

corepack pnpm install --frozen-lockfile
corepack pnpm --filter @promptflow/engine build
corepack pnpm --filter @promptflow/mcp-server build

Replace <ABSOLUTE_PATH_TO_CHECKOUT> below with your checkout. On Windows use forward slashes in JSON (C:/src/Prompt/…).

Configure your client

{
  "servers": {
    "promptflow": {
      "type": "stdio",
      "command": "node",
      "args": [
        "<ABSOLUTE_PATH_TO_CHECKOUT>/packages/mcp-server/dist/bin.js"
      ]
    }
  }
}

{
  "mcpServers": {
    "promptflow": {
      "command": "node",
      "args": [
        "<ABSOLUTE_PATH_TO_CHECKOUT>/packages/mcp-server/dist/bin.js"
      ]
    }
  }
}

claude mcp add promptflow -- node <ABSOLUTE_PATH_TO_CHECKOUT>/packages/mcp-server/dist/bin.js
claude mcp list

File locations follow each client's documentation; if a client moves them, its documentation wins. The server needs no secrets, so never add credentials to these files.

Verify

printf '%s\n' \
  '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"check","version":"1"}}}' \
  '{"jsonrpc":"2.0","method":"notifications/initialized"}' \
  '{"jsonrpc":"2.0","id":2,"method":"tools/list"}' \
  | node packages/mcp-server/dist/bin.js

Or run the TypeScript example, which starts the server, lists its tools, and evaluates and restructures a prompt:node --experimental-strip-types packages/mcp-server/examples/mcp-client.ts. In your client, ask “List the promptflow tools”.

Tools (15)

Arguments, results, and error cases for each tool: MCP reference. Example request in an agent chat:

Use promptflow to restructure this, then scan it for secrets:
"fix login issue and make google login work and don't break anything"

Troubleshooting

• Server shows as failed: the path is wrong or dist/ is not built. Rebuild and use the absolute path.
• node not found: the client has no Node.js on its PATH. Put the absolute path to node in command.
• Invalid arguments: the message names the field. Long input: set PROMPTFLOW_MCP_MAX_PROMPT_CHARS in the client's env.
• Need logs: set PROMPTFLOW_MCP_LOG=debug; they go to stderr and never contain prompt text.

Option 2: HTTP API

Base URL https://promptflowengine.com/api/v1. JSON in, JSON out, no authentication, nothing stored, 60 POST requests per minute per IP. Do not send prompts that contain secrets to any remote service; redact first with POST /sanitize, the CLI, or the extensions.

curl -s https://promptflowengine.com/api/v1/health

curl -s https://promptflowengine.com/api/v1/analyze \
  -H 'content-type: application/json' \
  -d '{"prompt":"Write something about our product and stuff. Keep it short but also be detailed."}' \
  | jq '{quality: .quality.overall, findings: [.findings[] | {severity, message}]}'

curl -s https://promptflowengine.com/api/v1/optimize \
  -H 'content-type: application/json' \
  -d '{"prompt":"Please summarize the notes. Please summarize the notes.","mode":"balanced"}' \
  | jq '{text: .selected.text, tokenDelta, changes: [.changes[].title]}'

import { createClient, PromptFlowApiError } from './api-client';

const api = createClient({ baseUrl: process.env.PROMPTFLOW_API ?? 'https://promptflowengine.com' });
try {
  const analysis = await api.analyze(userPrompt);
  const critical = analysis.findings.filter((f) => f.severity === 'critical');
  const { selected, tokenDelta } = await api.optimize(userPrompt, 'balanced');
  console.log(critical.length, selected.text, tokenDelta);
} catch (err) {
  // Typed errors carry status, type, message, and the request id for support.
  if (err instanceof PromptFlowApiError) console.error(err.status, err.type, err.message, err.requestId);
  else throw err;
}

Every endpoint, field, and status code: HTTP API reference.

Run it locally

corepack pnpm --filter @promptflow/web build
cd apps/web && corepack pnpm exec wrangler dev --local --env="" --port 8787
curl -s http://localhost:8787/api/v1/health

Troubleshooting

• 400: send content-type: application/json; the message names the field.
• 404 “Unknown API version”: use the /api/v1/ prefix.
• 405: health, models, and rules are GET; the rest are POST.
• 413: the body is over 256 KB or a field over 100,000 characters.
• 429: wait for retry-after (60 seconds), or run the engine locally with the CLI.