Evaluate state with typed choice, score, and probability questions.
io.github.pollinations/ask-jev (MCP Server)
The io.github.pollinations/ask-jev MCP server evaluates state using typed choice questions, plus scoring and probability questions. Its description indicates it focuses on structured evaluation rather than free-form input.
๐ ๏ธ Key Features
Typed choice questions for state evaluation
Score-based questions
Probability questions
๐ Use Cases
Assessing a state using predefined typed choices
Ranking or scoring answers as part of evaluation
Evaluating likelihood or probability-related responses
โก Developer Benefits
Clear question types (choice, score, probability) for structured workflows
Consistent evaluation inputs implied by โtyped choiceโ framing
โ ๏ธ Limitations
Available information does not specify tool names, interfaces, or supported question schemas beyond the three described categories
Nine models get one question that code can answer. See the anonymous vote, pick the answer you trust, then reveal who was right. 66 recorded rounds: the herd never lost. BYOP.
A game AI agents play over MCP: each picks one option in secret and scores for matching others. Connect any MCP agent, or seat Pollinations models with your own Pollen.
A tiny real computer written in code checks a model playing it: every step is diffed cell by cell, so you see where each model first slips. You pay with your own Pollen.
Put up to four Pollinations models on a small island with nothing to do. Plain code runs the world and counts what each one did; a timeline shows every turn. You pay with your own Pollen.
Three paragraphs on one topic: one by a person before LLMs, two by a model, one told to write plainly. Find the person, blind. A slop-score detector plays the same round against you. BYOP.
One AI keeps a sealed word, another gets 20 yes/no questions. Code judges the guess, a referee model re-checks every answer, a hash seals the word. Race the Asker. BYOP.
Name a science topic and a Pollinations model writes rival explanations, each with a prediction. You run observations and the explanations that fail are falsified. BYOP, you pay your own Pollen.
Turns a title, an author and one line of premise into a finished 1600x2400 book cover. Pollinations paints the artwork, the page sets the type, and you download the PNG.
Agents can find MCP serversNov 16 Use listMcpServers to discover other hosted MCP servers, including their tools, pricing, and alpha flags.
Inspect Docker Hub tagsNov 16 The Packages MCP server can list tags for a public Docker Hub image or look up an exact tag. Results include its digest, size, last push time, and supported platforms.
Exact math for agentsNov 16 The calculators MCP server gained integer_math, convert_number_base, and check_digit. Agents can work with large integers without number rounding, convert bases 2 through 36, and validate Luhn, IBAN, ISBN, and EAN-13 values.
Look up ASN registrationsNov 16 Use asn_whois in the Network MCP server to look up an ASN's registered range, organization, country, status, and other RDAP details.
See required model checksNov 16 OpenAI-compatible model listings now include required_safety, including mandatory checks inherited from fallbacks. Clients can read the checks without querying the registry separately; the checks themselves have not changed.
See per-use model chargesNov 16 Model listings now include pricing_adjustments when a model has per-use charges beyond token pricing, such as Gemini search queries or cache storage. Models without adjustments omit the field.
Copy docs from dashboard cardsNov 14 Open the relevant guide from a dashboard card, or copy its text into an AI chat. The Keys card picks docs for your key type.
Agents get their own pageNov 13 Browse Agents and MCP on dedicated pages. Manage your Agents, Models and Apps together in My listings.
Browse sandbox startersNov 13 Browse sandbox starter templates and prices without signing in. Sign in to create a sandbox or view yours.
Keys in two tabsNov 13 Switch between Secret and App keys in one card. Open the App tab directly with /keys?tab=app.
pollinations.ai is an open-source generative AI platform based in Berlin, used by 500+ community projects. It provides APIs for text, image, video, audio, 3D, and embeddings generation. Our code and development discussions are public. Supporters help fund the project.
๐ Key Features
๐ 100% Open Source: code, decisions, roadmap all public
๐ค Community-Built: 500+ projects already using our APIs
๐ฑ Quests: complete Quests to add to your balance
๐ผ๏ธ Image Generation: Text-to-image and image editing
๐ Text Generation: Chat, reasoning, vision, function calling, structured outputs
๐ฌ Video Generation: Text-to-video and image-to-video
๐ต Audio: Text-to-speech and speech-to-text
๐ง 3D Generation: Text-to-3D and image-to-3D
๐๏ธ Real-time API: OpenAI-compatible WebSocket for streaming conversations
Build a reusable text agent without hosting an agent server. Choose a system prompt and base model, optionally give it Pollinations tools, then call it through the same OpenAI-compatible API as any other model.
Private or public: keep an agent owner-only or publish it after receiving community publisher access.
Normal model API: call a registered agent by its owner/agent-name model ID.
Dashboard, CLI, or API: manage agents in My listings, with polli agents, or through /account/agents.
Run tools such as Bloom CLI, DeepSeek Harness, Hermes Agent, Mistral Vibe, OpenCode, Pi, Prime Agent, and tgpt on Pollinations models. polli harness edits the tool's own config so it calls Pollinations with a dedicated key, and restores it on off.
bash
npx @pollinations/cli harness dsh on
See Coding Harnesses for supported harnesses and how to add one.
๐งฉ Community Models
Community members connect their own OpenAI-compatible text, image, image-editing, and speech-to-text endpoints to Pollinations.
Host your own model: follow Publish a Model to test and register an upstream endpoint, then keep it private or publish it to the catalog.
Automatic fallback routing: nominate up to three compatible backup models so generations keep moving when an upstream model goes down.
Use duration to set video length, aspectRatio for orientation, and image[0]/image[1] to pass start/end reference frames. See available video models and capabilities at gen.pollinations.ai/video/models.
Pass reference image URL(s) via the image parameter for image-to-3D models (put image= last in the URL, or URL-encode it). See available 3D models at gen.pollinations.ai/3d/models.
Pass a string or an array of up to 32 strings (text or multimodal content parts). See available embedding models at gen.pollinations.ai/embeddings/models.
Pollinations CLI
Generate text, images, audio, video, and more right from your terminal:
bash
npx @pollinations/cli gen image "cyberpunk city at night" --model black-forest-labs/flux.1-schnell --output city.png
npx @pollinations/cli gen text "Explain quantum tunneling in one sentence"
npx @pollinations/cli gen audio "Hello world" --voice nova --output speech.mp3
npx @pollinations/cli gen video "a waterfall in slow motion" --duration 5 --output clip.mp4
npx @pollinations/cli gen transcribe speech.mp3
Install globally and use the shorter polli command: npm install -g @pollinations/cli. Run polli models to list models, polli auth login to authenticate, and polli docs for the full API reference in your terminal.
Real-time API
Stream AI responses over an OpenAI-compatible WebSocket:
Browser clients pass a user-authorized key from Pollinations Pay as a query parameter (?key=); server clients can use the Authorization: Bearer header instead.
MCP Servers
Connect Streamable HTTP clients to the hosted Pollinations MCP server:
text
https://gen.pollinations.ai/mcp/pollinations
Send your Pollinations API key as an Authorization: Bearer header. See the
MCP Servers documentation
for the full hosted catalog.
โ ๏ธ Never expose sk_ keys in client-side code, git repos, or public URLs
๐ก Building an app? Use Pollinations Pay: users pay for their own usage, you pay $0
Model Restrictions
Each API key can be scoped to model categories or specific models. When creating a key at enter.pollinations.ai, you can:
Allow all models: key works with any available model
Restrict to categories: e.g. only image, or image and text. A key allows every model in its categories, including models added later
Restrict to specific models: through the API or polli keys update <id> --models black-forest-labs/flux.1-schnell tongyi-mai/z-image-turbo; a model ID allows only that model
%%{init: {'theme': 'dark', 'themeVariables': { 'background': '#1a1a1a', 'primaryColor': '#2a2a2a', 'primaryBorderColor': '#555555', 'primaryTextColor': '#eeeeee', 'lineColor': '#00e5ff', 'clusterBkg': 'transparent', 'clusterBorder': '#888888', 'fontSize': '13px', 'fontFamily': 'Inter, system-ui, sans-serif'}}}%%
graph LR
subgraph CLIENTS["Clients / Apps"]
Q[Bots - Discord, Telegram, WhatsApp]
N[30+ Mobile and Web Apps]
A[pollinations.ai Web Frontend]
R[AI Agents - Qwen, Sillytavern, ...]
AI[AI Assistants - Claude]
MCP[MCP Server]
end
AI --> MCP
Q --> GEN
N --> GEN
A --> GEN
R --> GEN
MCP --> GEN
GEN["gen.pollinations.ai: Edge Router + Generation Worker"]:::cfWorker -->|auth and billing| ENTER["enter.pollinations.ai: Auth Gateway + Billing"]:::cfWorker
GEN --> IMG["Image: gen Worker dispatch to providers / GPU backends"]:::cfWorkerLight
IMG --> D["Flux, Z-Image, Seedream, ...: GPU VMs"]:::gpuNode
GEN --> TXT["Text: Portkey multi-provider"]:::provider
GEN --> VID["Video: Wan / Veo / Seedance"]:::provider
GEN --> AUD["Audio: ElevenLabs / OVH (Whisper)"]:::provider
style CLIENTS fill:none,stroke:#888,stroke-width:2px,stroke-dasharray: 5 5
linkStyle default stroke-width:3px,stroke:#00E5FF
classDef cfWorker fill:#E65100,color:#fff,stroke:#FFB300,stroke-width:2px,font-weight:bold
classDef cfWorkerLight fill:#BF360C,color:#fff,stroke:#FFB300,stroke-width:1px
classDef gpuNode fill:#064E3B,stroke:#34D399,color:#ECFDF5,stroke-width:2px
classDef provider fill:#1E3A8A,stroke:#60A5FA,color:#EFF6FF,stroke-width:1px
๐ฎ Future Developments
We're exploring:
Digital Twins: Creating interactive AI-driven avatars
Music Video Generation: Combining AI-generated visuals with music
Real-time AI-driven Visual Experiences: Projects like our Dreamachine
๐ Our Vision
We want AI tools to be:
Open & Accessible: Add to your balance by contributing, without a credit card
Transparent & Ethical: Inspect our code and discuss how the platform works
Community-Driven: Build tools with other developers, creators, and AI enthusiasts
Interconnected: Combine models and services through shared APIs
Evolving: Add new capabilities while keeping the platform open and accessible
๐ค Community and Development
You can contribute to pollinations.ai in several ways:
Coding Assistant: Create a GitHub issue describing a feature for the Polli assistant to work on. You can propose a change without writing code.
Project Submissions: Have you built something with pollinations.ai? Use our app submission template to share it with the community and get it featured in our README.
Feature Requests & Bug Reports: Have an idea or found a bug? Open an issue and let us know. Our team and the Polli assistant will review it.