AI Developer Tools
Prepare prompts, test model APIs, generate AI CLI commands, validate structured JSON output, and inspect live MCP tools in one workflow.
AI Prompt Toolkit
Redact secrets, estimate tokens, and split long prompts before you paste them into an external AI tool.
Sanitize and Split AI Prompts
Open the keyword-forward guide, then jump into safe prompt-toolkit presets for redaction, chunking, and cleanup.
Agent Skill Builder
Turn Markdown notes into a browser-local Claude or Codex SKILL.md file with deterministic frontmatter and copy-ready output.
CLI comparison guide
Compare Claude Code, Codex, and OpenCode command patterns before you choose a live preset.
AI CLI Tools
Generate copy-ready Claude Code, Codex, and OpenCode commands without jumping between docs tabs.
Structured Output Playground
Validate JSON-shaped model output against the contract your app expects before you wire it into code.
LLM API Tester
Call OpenAI-compatible APIs directly from your browser and inspect models, streaming output, headers, timing, and usage.
Structured Output JSON Validator
Open a keyword-forward guide with safe preset links into the live structured-output playground.
MCP Server Check
Connect to a remote Streamable HTTP server, inspect its tools, and run confirmed calls.
I need to clean a prompt before I share it.
Start with prompt cleanup, redaction, and chunking when the input includes repo notes, incident text, or secrets.
Open AI Prompt Toolkit
I need the exact command and flags for an AI coding assistant.
Use the CLI builder when you want the right command for Claude Code, Codex, or OpenCode without re-reading reference docs.
Open AI CLI Tools
I need to test an OpenAI-compatible endpoint.
Use a browser-direct profile to load models, send Chat or Responses requests, and inspect compatibility without a ComUtil proxy.
Open LLM API Tester
I need to validate JSON output before my app consumes it.
Use the structured-output workbench to catch missing fields, wrong types, or extra keys before you trust the payload.
Open Structured Output Playground
I need to verify an MCP server and try one of its tools.
Review the transport, required fields, and environment usage before you connect this config in an MCP client.
MCP Server Check
I need to turn notes into an agent skill.
Use the skill builder when a rough Markdown draft needs clean SKILL.md frontmatter and a copy-ready file.
Open Agent Skill Builder
Each step narrows what can still be wrong. Cleaning the prompt first means a later failure is not caused by text you never meant to send; testing the endpoint before validating output means a schema error is a model problem rather than a transport one.
1. Prompt prep
Clean and split the text before it leaves your workflow.
2. Command build
Generate the exact CLI invocation for the assistant you are using.
3. API test
Call the selected endpoint directly and inspect its real response.
4. Output check
Validate that the JSON response still satisfies your expected shape.
5. Config debugging
Review the transport, required fields, and environment usage before you connect this config in an MCP client.
Getting a model into an application is five separate jobs that usually happen in five separate places: making the prompt safe to send, building the command or request, calling the endpoint, checking that the answer still matches the shape your code parses, and confirming the server config an agent will load. Every page linked here does one of those jobs, and each one runs in the browser tab you already have open, so the prompt, key, or payload you are debugging does not have to be pasted into a service you do not control.
The order on this page is the order the work usually happens in. Prompt cleanup, CLI command building, skill file generation, and structured output validation never send your input anywhere: they run entirely in page scripts. Two of them have to reach the network, and both do it directly from your browser to the endpoint you named, with credentials kept in this browser's local storage. That direct path is the reason a failure here is worth trusting: when the LLM API tester or the MCP server check reports an error, it came from your endpoint and its transport, CORS, or auth configuration, not from a proxy in between.
- Stripping keys, hostnames, and customer text out of a prompt before it goes to an external assistant
- Getting the exact Claude Code, Codex, or OpenCode invocation instead of reading three sets of reference docs
- Confirming that a new base URL, key, or gateway answers at all, and how quickly, before it is wired into code
- Checking that a model's JSON still satisfies the contract your application parses
- Verifying an MCP server's transport and tool list before adding it to a client config
- Turning a rough Markdown draft into a SKILL.md file an agent can load
- Prompt toolkit, CLI builder, skill builder, structured output: input never leaves the page
- LLM API Tester: your browser calls the endpoint you configured, and profiles and keys stay in local storage
- MCP Server Check: connects to the server you name and asks for confirmation before it runs a tool call
- Guide pages carry no input fields; they explain a workflow and link into the live tool
- Nothing on this page uploads a prompt, key, or payload to ComUtil for processing
Where do I start if I only have a rough prompt?
Start with the AI Prompt Toolkit to redact secrets and split the text into chunks that fit a context window, then move to the CLI builder or the API tester depending on whether you are driving a coding assistant or calling an endpoint yourself.
Do any of these tools send my prompt or API key to ComUtil?
No. The text tools run entirely in the browser, and the two tools that make network calls send them from your browser straight to the endpoint you configured. Keys are stored in this browser's local storage, and clearing site data removes them.
What is the difference between the CLI builder and the LLM API tester?
The CLI builder produces command text for an AI coding assistant such as Claude Code, Codex, or OpenCode; it never runs anything. The API tester actually sends a request to a model endpoint and shows the status, streamed tokens, timing, and usage that came back.
Why does the API tester fail when the same request works from my terminal?
A browser request needs the provider to return CORS headers, and a terminal request does not. If the endpoint sends none, the browser blocks the response before the page can read it. Put a gateway that returns CORS headers in front of the model, or run the redacted curl command the tester generates.
Which page do I use to check an MCP server before adding it to a client?
MCP Server Check. It connects to a remote Streamable HTTP server, lists the tools it advertises with their input schemas, and runs a call only after you confirm it, which is enough to tell a reachable server from a misconfigured one.