Replicate Streamable
FreeNot checkedMCP Server for interacting with public models on Replicate. Written in TypeScript, Node and Hono.dev
About
MCP Server for interacting with public models on Replicate. Written in TypeScript, Node and Hono.dev
README
Lightweight MCP server for AI image generation and editing using Replicate's official models.
Release status (2026-07-27): this repository pins
@modelcontextprotocol/serverand the test-only@modelcontextprotocol/clientto2.0.0-beta.5, with Zod 4 and the candidate2026-07-28protocol. The dated protocol and stable v2 SDK are not final at this commit; do not claim final conformance until the release gate is verified.
The Bun and Cloudflare Workers entry points share one fetch-native handler per deployment and create a fresh MCP server for every request. Modern HTTP is stateless; compatibility with 2025-era clients uses the SDK's stateless fallback and does not create MCP sessions.
Author: overment
Overview
This is a focused, minimal MCP server designed specifically for image generation workflows. Unlike full-featured Replicate MCP Servers, this server exposes only what's needed for image generation and editing with official models like Flux, SDXL, and Seedream.
Recommended usage: Tell your assistant upfront which model you prefer and any specific settings (quality, aspect ratio, style). This reduces tool calls and gets better results faster.
Example system prompt addition:
For image generation, use black-forest-labs/flux-schnell with 16:9 aspect ratio unless I specify otherwise.
Notice
This repo works in two ways:
- As a fetch-native Bun server for local workflows
- As a fetch-native Cloudflare Worker for remote interactions
Features
- Search Models — Find image generation models with full input schemas
- Generate Images — Run predictions and get results with expiring URLs
- Lightweight — Only 2 tools
- API Key Auth — Simple Bearer token or X-Api-Key header authentication
- Dual Runtime — Node.js/Bun or Cloudflare Workers
Design Principles
- LLM-friendly: Two focused tools, not 1:1 API mirrors
- Schema-aware: Search returns input schemas so agent knows exact parameters
- Secure: Replicate API key stored as secret, clients use separate bearer token
- Clear feedback: Model parameters, generation time, markdown-ready output
Installation
Prerequisites: Bun, Replicate Account.
Ways to Run (Pick One)
- Local Development — Standard setup with bearer token auth
- Cloudflare Worker (wrangler dev) — Local Worker testing
- Cloudflare Worker (deploy) — Remote production
1. Local Development — Quick Start
Get Replicate API token:
- Visit replicate.com/account/api-tokens
- Create a new API token
- Copy the token (starts with
r8_)
Configure environment:
cd replicate-mcp
bun install
cp env.example .env
Edit .env:
PORT=3000
# Generate with: openssl rand -hex 32
API_KEY=your-random-auth-token
# Replicate API Token (from replicate.com)
REPLICATE_API_TOKEN=r8_your_token_here
- Run:
bun dev
# MCP: http://127.0.0.1:3000/mcp
Claude Desktop / Cursor:
{
"mcpServers": {
"replicate": {
"command": "npx",
"args": ["mcp-remote", "http://localhost:3000/mcp", "--transport", "http-only"],
"env": { "NO_PROXY": "127.0.0.1,localhost" }
}
}
}
2. Cloudflare Worker (Local Dev)
bun x wrangler dev --local | cat
Create .dev.vars for local secrets:
API_KEY=your_random_auth_token
REPLICATE_API_TOKEN=r8_your_replicate_token
Endpoint: http://127.0.0.1:8787/mcp
3. Cloudflare Worker (Deploy)
Update
wrangler.jsoncfor the production URL and exact Host/Origin allowlists. The checked-in values are local-safe defaults.Set secrets:
# Generate a random token for client authentication
openssl rand -hex 32
bun x wrangler secret put API_KEY
# Paste the generated token when prompted
# Replicate API token
bun x wrangler secret put REPLICATE_API_TOKEN
# Paste your Replicate token when prompted
- Validate generated types and deploy:
bun run types:worker
bun run types:worker:check
bun run build:worker
bun run deploy
Endpoint: https://<worker-name>.<account>.workers.dev/mcp
Client Configuration
Alice App
Add as MCP server with:
- URL:
https://your-worker.workers.dev/mcp - Type:
streamable-http - Header:
Authorization: Bearer <your-API_KEY>
Claude Desktop / Cursor (Local Server)
{
"mcpServers": {
"replicate": {
"command": "npx",
"args": ["mcp-remote", "http://localhost:3000/mcp", "--transport", "http-only"],
"env": { "NO_PROXY": "127.0.0.1,localhost" }
}
}
}
Claude Desktop / Cursor (Cloudflare Worker)
{
"mcpServers": {
"replicate": {
"command": "npx",
"args": ["mcp-remote", "https://your-worker.workers.dev/mcp", "--transport", "http-only"]
}
}
}
MCP Inspector (Quick Test)
bunx @modelcontextprotocol/inspector
# Connect to: http://localhost:3000/mcp (local) or https://your-worker.workers.dev/mcp (remote)
Tools
search_models
Search for models and get their input schemas. Returns up to 5 models with full parameter details.
// Input
{
query: string; // Model name, task, or keywords
}
// Output
### owner/name
Description of the model
Runs: 1,234,567
Input parameters:
- prompt [REQUIRED]: string
- aspect_ratio: enum: ["1:1", "16:9", "9:16"] = "1:1"
- num_outputs: integer = 1
...
Example:
{ "query": "flux" }
generate_image
Run an image generation model and wait for the result.
// Input
{
model: string; // "owner/name" format
input: Record<string, unknown>; // Model-specific parameters
}
// Output
## Image Generated in 2.3s
Model: black-forest-labs/flux-schnell
Display the image to the user using markdown syntax:

Note: URLs expire in 1 hour.
Common input patterns:
// Text-to-image
{
"model": "black-forest-labs/flux-schnell",
"input": {
"prompt": "a cat on the moon",
"aspect_ratio": "16:9"
}
}
// Image editing
{
"model": "black-forest-labs/flux-kontext-pro",
"input": {
"prompt": "change the sky to sunset",
"image": "https://example.com/source.jpg"
}
}
Popular Models
| Model | Speed | Best For |
|---|---|---|
black-forest-labs/flux-schnell |
~2s | Quick generations, drafts |
black-forest-labs/flux-dev |
~10s | Higher quality, detailed |
bytedance/seedream-4 |
~5s | Versatile, multi-reference |
black-forest-labs/flux-kontext-pro |
~8s | Image editing with text |
Aspect Ratio Guide
| Ratio | Use Case |
|---|---|
1:1 |
Portraits, icons, profile pictures |
16:9 |
Landscapes, cinematic, desktop wallpapers |
9:16 |
Mobile wallpapers, stories, vertical content |
4:3 |
Classic photo format |
21:9 |
Ultra-wide cinematic |
HTTP Endpoints
| Endpoint | Method | Purpose |
|---|---|---|
/mcp |
POST | MCP JSON-RPC 2.0 |
/health |
GET | Health check |
Environment Variables
Node.js (.env)
| Variable | Required | Description |
|---|---|---|
REPLICATE_API_TOKEN |
✓ | Replicate API token (r8_...) |
API_KEY |
✓ | Auth token for MCP clients |
PORT |
Server port (default: 3000) | |
HOST |
Server host (default: 127.0.0.1) |
Cloudflare Workers (wrangler.jsonc + secrets)
Relevant wrangler.jsonc vars:
"vars": {
"MCP_TITLE": "Replicate",
"MCP_VERSION": "1.0.0"
}
Secrets (set via wrangler secret put):
API_KEY— Random auth token for clientsREPLICATE_API_TOKEN— Replicate API token
No KV namespace or MCP session store is required by the v2 serving shell.
Development
bun dev # Start with hot reload
bun run typecheck # TypeScript check
bun run lint # Lint code
bun run build # Bun production build
bun run build:worker
bun run types:worker:check
bun test # Modern, legacy, security, and provider-mock tests
bun start # Run Bun production entry point
Architecture
src/
├── config/
│ └── metadata.ts # Tool descriptions
├── tools/
│ ├── search-models.tool.ts # Search with schema enrichment
│ └── generate-image.tool.ts # Run predictions
├── services/
│ └── api/
│ └── replicate.service.ts # Replicate API client
├── http/
│ ├── app.ts # Fetch-native HTTP shell
│ ├── auth.ts # MCP API-key boundary
│ ├── body.ts # Bounded request bodies
│ └── security.ts # Host, Origin, and strict CORS
├── core/
│ ├── mcp.ts # Fresh server factory
│ └── runtime.ts # Deployment-scoped v2 handler
├── index.ts # Bun entry
└── worker.ts # Workers isolate entry
Troubleshooting
| Issue | Solution |
|---|---|
| 401 Unauthorized | Check API_KEY is set and client sends Authorization: Bearer <token> |
| "REPLICATE_API_TOKEN not configured" | Set secret: wrangler secret put REPLICATE_API_TOKEN |
| "Invalid model format" | Use owner/name format (e.g., black-forest-labs/flux-schnell) |
| "Missing required parameters" | Call search_models to see exact input schema |
| "Rate limit exceeded" | Wait a moment and retry |
| "Image URL expired" | URLs expire after 1 hour — generate again |
Debugging
Test with MCP Inspector:
bunx @modelcontextprotocol/inspector
# Connect to your endpoint and test tools
Check Worker logs:
wrangler tail
License
MIT
Installing Replicate Streamable
This server has no published package — it is built from source. Open the repository and follow its README.
▸ github.com/iceener/replicate-streamable-mcp-serverFAQ
Is Replicate Streamable MCP free?
Yes, Replicate Streamable MCP is free — one-click install via Unyly at no cost.
Does Replicate Streamable need an API key?
No, Replicate Streamable runs without API keys or environment variables.
Is Replicate Streamable hosted or self-hosted?
Self-hosted: the server runs locally on your machine via the install command above.
How do I install Replicate Streamable in Claude Desktop, Claude Code or Cursor?
Open Replicate Streamable on unyly.org, pick your client tab (Claude Desktop, Claude Code, Cursor) and press Install — the config is generated automatically, no JSON editing.
Related MCPs
GitHub
PRs, issues, code search, CI status
by GitHubFilesystem
Secure file operations with configurable access controls.
Memory
Knowledge graph-based persistent memory system.
Template MCP Server
A CLI tool to create a new Model Context Protocol server project with TypeScript support, dual transport options, and an extensible structure
by mcpdotdirectCompare Replicate Streamable with
Not sure what to pick?
Find your stack in 60 seconds
Author?
Embed badge for your README
Browse similar
All development MCPs
