The Stack · LLM gateways & inference API

Replicate

Run thousands of community and official AI models (image, video, audio, LLM) via one HTTP API with no ML infra to manage.

API keyUsage-based🔵 agent: oauthMCP ✓

Get API key → Pay per second of GPU time or per output (e.g. FLUX ~$0.003-$0.04/image); no subscription, idle/setup time free Docs Website

CategoryLLM gateways & inference

TypeAPI service

Authapi-key

PricingUsage-based

Agent signup🔵 oauth

SDKsPython, TypeScript, REST, MCP

MCP serveravailable →

Pricing: Pay per second of GPU time or per output (e.g. FLUX ~$0.003-$0.04/image); no subscription, idle/setup time free

🔵 Agents: oauth

OAuth — connect the account once, then agents act on its behalf.

Account creation is GitHub/Google OAuth (human step); a default API token is issued on account creation; billing/card needed for most models beyond free runs

Replicate — Run thousands of community and official AI models (image, video, audio, LLM) via one HTTP API with no ML infra to manage.
Website: https://replicate.com
Docs: https://replicate.com/docs
Get a key: https://replicate.com/account/api-tokens
Agent signup: oauth — Account creation is GitHub/Google OAuth (human step); a default API token is issued on account creation; billing/card needed for most models beyond free runs
Auth: api-key
MCP server: https://mcp.replicate.com
Pricing: usage-based — Pay per second of GPU time or per output (e.g. FLUX ~$0.003-$0.04/image); no subscription, idle/setup time free
Full record: https://dreaming.press/api/tools/replicate.json

Machine record: /api/tools/replicate.json · add its MCP server →

Quickstart

import replicate
out = replicate.run(
    "black-forest-labs/flux-schnell",
    input={"prompt": "an astronaut riding a horse"})
print(out)

python · full docs →

Model Context Protocol

Replicate exposes an MCP server, so you can add its tools to Claude, Cursor, or any MCP client: https://mcp.replicate.com

Replicate in our coverage

Milvus Replaced Kafka and Pulsar With a WAL on S3: Why the Write Path Was the Real Bottleneck

2026-07-08

Replicate FAQ

What is Replicate?

Replicate is Run thousands of community and official AI models (image, video, audio, LLM) via one HTTP API with no ML infra to manage. It's in the LLM gateways & inference category of the dreaming.press tool directory.

Is Replicate free?

Replicate is usage-based (you pay for what you use). Pay per second of GPU time or per output (e.g. FLUX ~$0.003-$0.04/image); no subscription, idle/setup time free.

Can an AI agent sign up for Replicate automatically?

Via OAuth — connect the account once and agents can then act on its behalf. Account creation is GitHub/Google OAuth (human step); a default API token is issued on account creation; billing/card needed for most models beyond free runs.

Does Replicate have an MCP server?

Yes — Replicate offers a Model Context Protocol server, so you can add its tools to Claude, Cursor, or any MCP client: https://mcp.replicate.com

Is Replicate open source?

Replicate is a hosted API service, not an open-source project.

What languages does Replicate support?

Replicate offers Python, TypeScript, REST, MCP.

Replicate

Quickstart

Model Context Protocol

Replicate in our coverage

Milvus Replaced Kafka and Pulsar With a WAL on S3: Why the Write Path Was the Real Bottleneck

Making a Pydantic AI Agent Crash-Proof: Temporal vs DBOS vs Prefect vs Restate

Claude Science's Reviewer Agent: How to Make Multi-Agent Output Reproducible

How to Read a Launch Benchmark When the Vendor Scored Its Own Exam

EPLB vs LPLB: Why SGLang's 5x MoE Speedup Was a Solver, Not a GPU

Does Structured Output Hurt LLM Accuracy? The Format Tax, Measured

Replicate FAQ

New agent tools & APIs, the week they ship