Indie Alternatives MCP Server

Indie Alternatives

English | 中文 · indiealternatives.dev · List your product

Indie-made alternatives to big-name software, inside any AI assistant:

1. When a user asks "is there a tool that does X?" or "what's an alternative to Typeform?", the AI searches a catalog of products made by indie makers and recommends the best matches.
2. When nothing fits, the need is logged anonymously, so indie makers can see what people want that doesn't exist yet.

Quick start

Add this MCP server URL to your AI assistant. No install, no API key:

https://indiealternatives.dev/mcp

Then ask: "What's a good alternative to Typeform?" or "Is there a tool that turns screenshots into nice social images?"

Makers: list your product for free.

All integration styles

One codebase, every integration style:

https://indiealternatives.dev/llms.txt describes the service for AIs that read web pages.

Tools

Every tool can also be called with POST /api/call/{tool_name}, with the arguments as the request body.

The service tells the AI how to use it (MCP instructions, OpenAPI info.description, llms.txt): when to search, when to log, and to strip personal info before logging.

Privacy

• A record contains only: a random id, the need text, optional domain / platform / budget, the date (day only), and a fingerprint used to group similar needs.
• Never stored: user identity, conversation content, IP addresses, User-Agent, exact timestamps. The HTTP server keeps no access log.
• Counting people, not requests: each need carries a short code = hash(random salt of the day + IP). The salt is deleted when the day ends, so codes can't be reversed or linked across days. The same person on the same day counts once on the public board. Need logging is also rate-limited at the edge (10 per minute per address; nothing stored).
• The server redacts again before writing: emails, phone numbers, links, IPs, ID and card numbers, @handles, messenger IDs and "my name is …" are replaced with placeholders like [email] and [phone].
• The tool description asks the AI to rewrite the need generically first, and not to log if the user declines.
• Local installs (npx indie-alternatives-mcp, the CLI) read the latest catalog from https://indiealternatives.dev/catalog.json (falling back to the bundled copy) and send unmet needs to https://indiealternatives.dev/api/needs, so needs from every install are pooled for makers. Only the redacted need fields are sent. Set INDIE_ALT_OFFLINE=1 to keep everything on your machine.
• Usage stats on the hosted service are daily counters only (e.g. "searches today", "searches with no match", "connections from client X"). Nothing per user or per request is stored.
• Makers' optional contact email from the submission form is kept private and used only for the review.

Install

Node.js 18+.

npm install
npm run build
npm test

1. Local MCP clients (run it yourself)

Most people should use the hosted URL above. To run locally from a clone (npm install && npm run build), almost every MCP client uses the same config, in a different file. The npm package (npx -y indie-alternatives-mcp) is coming soon:

{
  "mcpServers": {
    "indie-alternatives": {
      "command": "node",
      "args": ["/absolute/path/indie-alternatives/dist/src/stdio.js"]
    }
  }
}

(Once the npm package is out: "command": "npx", "args": ["-y", "indie-alternatives-mcp"].)

Config locations change often; check each client's docs.

To collect needs from everyone's local installs in one place, add "env": { "NEEDS_WEBHOOK_URL": "https://indiealternatives.dev/api/needs" }.

2. Deploy the remote server

Cloudflare Workers + D1 (what indiealternatives.dev runs on)

npx wrangler login
npx wrangler d1 create indie-alternatives          # copy the database_id into wrangler.jsonc
npm run db:migrate                                  # create the tables in D1
npx wrangler secret put ADMIN_TOKEN                 # a long random string, for the review API
npm run deploy

Workers Logs are disabled in wrangler.jsonc to keep the no-access-log promise. To serve on your own domain, add it to Cloudflare DNS and uncomment routes in wrangler.jsonc.

Local development, no Cloudflare account needed: npm run db:migrate:local && npm run dev:worker (put ADMIN_TOKEN=… in .dev.vars).

Node / Docker (self-hosting)

PORT=8787 PUBLIC_URL=https://your-domain DATABASE_FILE=./data.db ADMIN_TOKEN=… npm run start:http

DATABASE_FILE (SQLite via node:sqlite, Node 22.5+) enables maker submissions, the review API and stats; without it, needs go to a JSONL file. Or with Docker:

docker build -t indie-alternatives .
docker run -d -p 8787:8787 -e PUBLIC_URL=https://your-domain -e ADMIN_TOKEN=… -v indie-alternatives-data:/data indie-alternatives

For local testing, ngrok http 8787 or cloudflared tunnel --url http://localhost:8787 gives you a temporary HTTPS URL.

/healthz should return {"ok": true}.

Makers: submissions and review

Makers submit at /submit (HTML form) or POST /api/submissions (JSON: name, url, tagline, optional description, categories, tags, pricing, platforms, maker, contact). Submissions wait for review; duplicates and already-listed URLs are rejected. After approval the product is searchable immediately, and makers can show the badge at /badge.svg.

Review API (Authorization: Bearer $ADMIN_TOKEN):

curl -H "authorization: Bearer $ADMIN_TOKEN" https://indiealternatives.dev/api/admin/submissions            # ?status=pending|approved|rejected|all
curl -X POST -H "authorization: Bearer $ADMIN_TOKEN" -H "content-type: application/json" \
     -d '{"tags":["typeform alternative","online form"]}' https://indiealternatives.dev/api/admin/submissions/<id>/approve   # body: optional field fixes
curl -X POST -H "authorization: Bearer $ADMIN_TOKEN" -d '{"note":"not indie"}' https://indiealternatives.dev/api/admin/submissions/<id>/reject
curl -X DELETE -H "authorization: Bearer $ADMIN_TOKEN" https://indiealternatives.dev/api/admin/products/<id>
curl -H "authorization: Bearer $ADMIN_TOKEN" "https://indiealternatives.dev/api/admin/stats?days=30"       # anonymous daily counters
curl -H "authorization: Bearer $ADMIN_TOKEN" https://indiealternatives.dev/api/admin/needs                  # all unmet needs, grouped

The home page shows the most requested unmet needs publicly, but only needs asked at least 3 times.

3. Remote MCP clients

4. Platforms without MCP: import OpenAPI

Custom GPT (GPT Actions): Create a GPT → Configure → Create new action → Import from URL: https://indiealternatives.dev/openapi.json. Auth "None" (or "API Key / Bearer" if you set MCP_AUTH_TOKEN). Add to the GPT instructions:

When the user asks whether there is a tool for something, call search_tools first. If nothing fits, call log_unmet_need to record the need anonymously, after removing all personal information.

Dify: Tools → Custom → Create custom tool → import https://indiealternatives.dev/openapi.json.

n8n / Zapier / others: call the REST endpoints with an HTTP node, or import the OpenAPI spec.

Gemini (Vertex AI Agent Builder): add an OpenAPI tool with the same URL.

5. Your own agent: function calling definitions

curl https://indiealternatives.dev/functions.json?format=openai            # OpenAI Chat Completions tools
curl https://indiealternatives.dev/functions.json?format=openai-responses  # OpenAI Responses API tools
curl https://indiealternatives.dev/functions.json?format=anthropic         # Claude Messages API tools
curl https://indiealternatives.dev/functions.json?format=gemini            # Gemini functionDeclarations

These also work with any OpenAI-compatible model API. When the model returns a tool call, POST the arguments to https://indiealternatives.dev/api/call/{tool_name} and hand the result back to the model.

Example (Python, OpenAI format):

import json, requests
from openai import OpenAI

BASE = "https://indiealternatives.dev"
tools = requests.get(f"{BASE}/functions.json?format=openai").json()
client = OpenAI()
messages = [{"role": "user", "content": "Is there a tool that adds nice backgrounds to screenshots?"}]

while True:
    msg = client.chat.completions.create(model="gpt-5", messages=messages, tools=tools).choices[0].message
    messages.append(msg)
    if not msg.tool_calls:
        print(msg.content)
        break
    for call in msg.tool_calls:
        result = requests.post(f"{BASE}/api/call/{call.function.name}", json=json.loads(call.function.arguments)).json()
        messages.append({"role": "tool", "tool_call_id": call.id, "content": json.dumps(result)})

6. CLI

npm link    # optional: install indie-alternatives as a global command
indie-alternatives search "add a nice background to screenshots"
indie-alternatives details shottr
indie-alternatives log "turn meeting recordings into notes with action items"
indie-alternatives needs

Output is JSON, so any agent that runs shell commands can use it.

Configuration (environment variables)

Bring your own catalog

data/products.json is the catalog served at indiealternatives.dev (1,500+ products). To run your own directory, replace it with an array (or { "products": [...] }) of:

{
  "id": "unique-id",
  "name": "Product name",
  "url": "https://...",
  "tagline": "One-line pitch",
  "description": "Longer description",
  "categories": ["Category"],
  "tags": ["keywords drive matching", "add synonyms", "in any language"],
  "pricing": "free | freemium | paid | commission",
  "platforms": ["web", "macos", "windows", "linux", "ios", "android", "self-hosted"],
  "maker": "optional: who made it"
}

To use a database or another API, implement CatalogSource in src/types.ts (a single load() method). Same for need storage: implement NeedStore (append() / list()), e.g. with Postgres or Supabase.

NEEDS_WEBHOOK_URL receives one record per request:

{ "id": "…", "need": "auto-generate subtitles for videos", "category": "Video", "day": "2026-11-05", "fingerprint": "373ca764a387d209" }

How matching works

No external services, everything runs locally: Latin-script text is split into words, CJK text into bigrams, filler words ("is there a tool", "recommend") are dropped, and products are scored with weights name > tags > tagline > categories > description. matched requires at least 60% of the meaningful part of the query to be covered. For a large catalog, swap src/search.ts for vector search; the interface stays the same.

Layout

src/
  core.ts      tool definitions and implementation (shared by every integration)
  server.ts    MCP server
  stdio.ts     local MCP entry
  app.ts       HTTP app on web-standard Request/Response: /mcp, /api/*, docs, pages, submissions, admin
  http.ts      Node HTTP entry (adapts node:http to app.ts)
  worker.ts    Cloudflare Workers entry (D1 storage)
  pages.ts     home page, submission form, badge
  rest.ts      OpenAPI, function calling and llms.txt generation
  catalog.ts   runtime-neutral catalog helpers
  sql.ts       D1 / SQLite stores (needs, stats, submissions, products)
  sqlite-node.ts  node:sqlite adapter
  cli.ts       CLI entry
  search.ts    scoring
  text.ts      tokenizer (Latin + CJK)
  privacy.ts   redaction and need fingerprints
  sources.ts   catalog sources and need stores (file, URL, webhook)
migrations/          D1 schema
wrangler.jsonc       Cloudflare Workers config
data/products.json   sample catalog
test/                tests
Dockerfile           container build

License

MIT