PromptingEasy — free tools for AI agents

Free to use. No account, API key, registration, payment or AI provider subscription is required to generate prompt text on PromptingEasy. Third-party AI platforms may have their own access requirements. The tools here generate prompt instructions; they do not run ChatGPT, Claude, Gemini, DeepSeek or other models, submit messages to them, or process payments.

Discover the service

All ten active prompt generators

One MCP server for every generator: connect to https://promptingeasy.com/mcp, call list_generators to choose a generator and language, call get_generator_details to learn its exact fields, then call generate_prompt. No separate MCP URL is needed for travel, cooking or any other topic. The REST and MCP interfaces support the same ten active generator IDs and 22 published locales.

Examples: Travel generator parameters (German) and Cooking generator parameters (German). Always read the exact options before sending a generate_prompt request; each generator has its own parameters.

Browser WebMCP: the matching generator tool is exposed on the generator's own page. A browser agent can navigate to the relevant generator through the generator directory. A remote MCP client can select any generator without navigating between pages. Tool availability depends on the host AI application connecting or enabling the interface.

REST API — example

First obtain the exact dropdown values with GET /api/v1/generators/writing-email?locale=en, then POST structured choices. Fields marked customAllowed also accept one free-text entry (maximum 600 characters). Fields marked multiple accept an array of dropdown selections plus at most one custom entry. Unknown field names and retired generators are rejected.

curl -s https://promptingeasy.com/api/v1/generators/writing-email?locale=en

curl -s https://promptingeasy.com/api/v1/prompts \
  -H 'Content-Type: application/json' \
  --data '{"generator":"writing-email","locale":"en","fields":{"task":"Write from scratch"}}'

Response fields include prompt and copyPayload. The default presentation: "classic" corresponds to the initial Step-4 copy text used by the website; an edited browser preview remains independent. For any of the 22 supported languages, set presentation: "guided" to add actionable sentences and missing-information placeholders without changing the browser Step-4 generator or Copy button. The ai-image prompt stays in its classic image format. The guided instructions are localized for all 22 supported languages; classic remains available everywhere.

Missing destination and appointment context: The travel generator can either help choose a destination or plan a trip, so making a city mandatory for every request would break destination-discovery tasks. When planning an existing trip, pass the city in source_text, e.g. Reiseziel: Wien; for emails include the current appointment, suggested alternative and recipient where known. The API does not invent details or send messages. In guided mode, unknown details are marked explicitly and the cooking role avoids implying that ingredients are already provided.

AI model compatibility

OpenAPI is provider-neutral. ChatGPT/OpenAI tools, Anthropic Claude tool use, Google Gemini function calling and DeepSeek function calling can use the documented REST operations through a suitable host application or HTTP adapter. The remote MCP endpoint supports the July 2026 stateless MCP revision and a stateless legacy 2025-11-25 handshake. Local compatibility tests are not a substitute for a live third-party connector test. An AI model does not automatically discover or invoke a tool merely because this file exists: the model's host must enable or connect that tool.

Gemini integration: Google now documents custom MCP Connected Apps and Gemini Developer Mode. If your Gemini account supports custom connected apps, explicitly add this MCP URL and perform a real tool call. A regular Gemini chat without the connector is not an MCP compatibility test. Publication in Google's Connected Apps directory is a separate application/review process; see Google's official MCP developer guide. CORS is not automatically required to be unrestricted for all browser or server-to-server clients.

WebMCP in your browser

Published generator forms also contain experimental WebMCP annotations for compatible browser agents. Chrome's official Origin Trial enrollment is handled separately; this site does not claim WebMCP is enabled until a valid origin-specific token is deployed and the real browser test succeeds. The Step-4 read-only WebMCP tool returns the current preview including user edits. The ordinary Step-4 Copy button remains available without WebMCP.

MCP tools

list_generators, get_generator_details, and generate_prompt. For MCP 2026-07-28, every request requires matching MCP-Protocol-Version, Mcp-Method and (for tool calls) Mcp-Name headers plus JSON-RPC _meta. Clients using 2025-11-25 first send initialize followed by notifications/initialized; subsequent POST requests use MCP-Protocol-Version: 2025-11-25 without the new method/name headers. This legacy implementation is stateless and does not issue session IDs or offer a GET/SSE push channel. Each tool advertises an outputSchema for successful structured results. No login or OAuth token is necessary for these deterministic tools.

Practical connector workflow: In Claude, Gemini or any other client that supports connecting a remote MCP server, explicitly add https://promptingeasy.com/mcp and enable the connector. Then call list_generators, get_generator_details, and generate_prompt in order. MCP is not automatically activated simply by pasting the website address into a chat. A browser GET to /mcp normally returns 405, not a health check. The Mcp-Method and Mcp-Name headers are required by the 2026-07-28 transport standard; older 2025-11-25 clients omit them.

Tool safety declarations: all three tools publish a descriptive title plus MCP annotations readOnlyHint: true, destructiveHint: false, idempotentHint: true, and openWorldHint: false. They return generator metadata or prompt text without sending emails, making bookings, calling AI models or changing external accounts. The prompt-creation operation processes the text supplied by the caller but does not publish or execute it. These annotations are descriptive hints, not a substitute for enforcing server-side security.

MCP registry and compatibility

Official MCP Registry identity: com.promptingeasy/prompt-generators (publication pending). The remote server manifest describes the endpoint https://promptingeasy.com/mcp. Registry listing and acceptance into Anthropic's separate Software Directory are independent processes. The endpoint has local protocol tests for both the July 2026 revision and the November 2025 legacy handshake; Claude.ai users have reported successful live tool calls, but this documentation does not attest to raw wire-level traces or acceptance into any official directory.

Three example agent requests

Troubleshooting

Opening /mcp in a browser with GET normally returns HTTP 405; clients must send POST requests. Modern clients need the matching version, method, name and metadata; legacy 2025-11-25 clients need the initialization exchange and version header. If a third-party connector still fails, record its exact protocol version, HTTP status and response instead of assuming compatibility. If a tool rejects generator fields, first call get_generator_details for the exact options and custom-text rules. Report reproducible failures through the support contact below.

Support and security contact

For integration support, operational incidents, security reports or tool concerns, email ask@promptingeasy.com. See our privacy policy and imprint for additional operator information.

Privacy, safety and usage

Browser mode: Standard prompt generation and copying happen locally. API or MCP mode: requests, including any chosen text fields, reach Cloudflare for transient processing and are returned to the caller. The new handlers do not intentionally store submitted prompt text or forward it to an AI company. Network providers may process ordinary HTTP metadata and service analytics according to their applicable policies. Do not send passwords, payment details, medical records or other sensitive material. Respect reasonable request volumes; access may be rate-limited for abuse prevention.

All public website pages remain crawlable, including for AI training, subject to applicable law and third-party content rights. There is no agent-only hidden content or special Google AI ranking markup here.

Operational status

The generator directory, API discovery links and browser WebMCP functionality have been checked on the production website. The MCP endpoint is POST-only; opening /mcp with a browser GET request returns 405 Method Not Allowed by design and is not a full functional check. Local MCP protocol and generator tests have passed. A Claude.ai user reported ten successful connector calls (including expected validation errors), but independent wire-level testing, other clients, and official directory publication remain outstanding. Live compatibility with third-party MCP clients is confirmed only through the user-reported Claude.ai tool calls, not an independent test. Use the MCP server/discover, tools/list and tools/call requests with the documented headers for the live test.