{"slug": "we-added-mcp-to-promptot-manage-production-prompts-from-claude-cursor-and-codex", "title": "We Added MCP to PromptOT: Manage Production Prompts from Claude, Cursor, and Codex", "summary": "PromptOT, a prompt management platform for production LLM applications, has added a Model Context Protocol (MCP) server to let AI coding assistants like Claude, Cursor, and Codex manage prompts directly from their interfaces. The MCP server exposes 23 tools across five areas, enabling assistants to list, edit, version, test, and publish prompts without leaving the coding environment. This integration aims to eliminate the context switch between AI tools and a separate dashboard, streamlining prompt management for development teams.", "body_md": "Developers increasingly use Claude, Cursor and Codex to write, review and improve application code.\n\nBut when those applications depend on production LLM prompts, there is usually a disconnect.\n\nYour AI coding assistant can modify the application, but it cannot safely:\n\nThe developer must leave the AI tool, open a separate dashboard, locate the prompt and repeat the change manually.\n\nWe wanted to remove that context switch.\n\nThat is why we added a Model Context Protocol server to [PromptOT](https://www.promptot.com).\n\nPromptOT MCP lets compatible AI assistants work directly with your PromptOT workspace through structured tools and scoped permissions.\n\nPromptOT is a prompt management platform for production LLM applications.\n\nInstead of keeping system prompts as hardcoded strings, PromptOT lets teams manage them as structured assets.\n\nA prompt can contain separate blocks for:\n\nTeams can version these prompts, create test cases, run evaluations and publish approved versions.\n\nApplications retrieve the compiled prompt through an API, allowing prompts to be updated without redeploying the application.\n\nThe dashboard remains useful for visual editing and review. But developers increasingly work inside AI-assisted tools, so we wanted PromptOT to be accessible there as well.\n\nThe Model Context Protocol gives AI clients a standard way to communicate with external tools and data sources.\n\nWithout MCP, connecting an assistant to PromptOT would require a custom integration for every client.\n\nWe would need one integration for Claude, another for Cursor, another for Codex and potentially another for every future AI tool.\n\nWith MCP, PromptOT exposes one structured tool catalog.\n\nAny compatible client can discover those tools, understand their input schemas and call them when the user makes a natural-language request.\n\nThe flow looks like this:\n\n```\nDeveloper\n   ↓\nClaude, Cursor or Codex\n   ↓\nPromptOT MCP server\n   ↓\nAuthenticated PromptOT API\n   ↓\nPrompt workspace\n```\n\nThe AI assistant does not receive unrestricted database access. It can only call the PromptOT operations exposed through its authorized MCP connection.\n\nThe PromptOT MCP server currently exposes 23 tools across five areas.\n\nThe assistant can:\n\nExample request:\n\n```\nList all prompts in my PromptOT workspace related to customer support.\n```\n\nThe assistant can work with the structured sections inside a prompt:\n\nFor example:\n\n```\nOpen the customer-support prompt and add a guardrail that\nprevents the assistant from requesting passwords or complete\npayment information.\n```\n\nThe AI client can find the prompt, inspect its current blocks and make the appropriate structured change.\n\nPromptOT supports runtime variables such as:\n\n```\n{{user_name}}\n{{plan}}\n{{tone}}\n```\n\nThrough MCP, an assistant can:\n\nExample:\n\n```\nAdd a tone variable to the onboarding-assistant prompt with\n“friendly” as its default value.\n```\n\nPrompt changes should not silently replace production behaviour.\n\nPromptOT MCP includes tools to:\n\nA developer can ask:\n\n```\nSave the current customer-support prompt as a new draft version.\nUse “Added billing escalation guardrail” as the changelog note.\n```\n\nOr:\n\n```\nCompare the current draft with the published version and\nsummarize the behavioural differences.\n```\n\nAI assistants can also manage PromptOT test cases.\n\nAvailable operations include:\n\nFor example:\n\n```\nInspect the support-agent prompt and create three test cases:\n\n1. A normal product question\n2. A billing escalation\n3. An off-topic request\n\nCheck the existing test cases first and do not create duplicates.\n```\n\nThe cases are saved to the same PromptOT prompt and become available to the rest of the team.\n\nSuppose we have a prompt named `support-agent`\n\n.\n\nIt contains a Role block and an Instructions block, but it does not have a strong guardrail for billing disputes.\n\nInside Claude, we can ask:\n\n```\nUsing PromptOT, inspect the support-agent prompt.\n\nAdd a guardrail requiring refund requests and duplicate-charge\nreports to be escalated to a human. Do not change any other\nprompt blocks.\n\nSave the result as a new draft version with a clear changelog note.\n```\n\nClaude can use PromptOT MCP to:\n\nThe same workflow can be performed from Cursor or Codex.\n\nThe prompt remains inside PromptOT, with its structure and history preserved.\n\nPromptOT supports local and hosted MCP connections.\n\nThe easiest Claude Desktop installation is the PromptOT Desktop Extension.\n\nDownload it here:\n\n[Download PromptOT for Claude Desktop](https://www.promptot.com/downloads/promptot.mcpb)\n\nOpen the downloaded file, and Claude Desktop will guide you through the installation.\n\nA manual configuration is also available:\n\n```\n{\n  \"mcpServers\": {\n    \"promptot\": {\n      \"command\": \"npx\",\n      \"args\": [\"-y\", \"@prompt-ot/mcp\"],\n      \"env\": {\n        \"PROMPTOT_API_KEY\": \"your_promptot_mcp_key\",\n        \"PROMPTOT_MCP_CLIENT\": \"claude-desktop\"\n      }\n    }\n  }\n}\n```\n\nAdd PromptOT to your Cursor MCP configuration:\n\n```\n{\n  \"mcpServers\": {\n    \"promptot\": {\n      \"command\": \"npx\",\n      \"args\": [\"-y\", \"@prompt-ot/mcp\"],\n      \"env\": {\n        \"PROMPTOT_API_KEY\": \"your_promptot_mcp_key\",\n        \"PROMPTOT_MCP_CLIENT\": \"cursor\"\n      }\n    }\n  }\n}\n```\n\nAdd PromptOT to the Codex configuration:\n\n```\n[mcp_servers.promptot]\ncommand = \"npx\"\nargs = [\"-y\", \"@prompt-ot/mcp\"]\nenv = { PROMPTOT_API_KEY = \"your_promptot_mcp_key\", PROMPTOT_MCP_CLIENT = \"codex-cli\" }\n```\n\nThe MCP package can also be launched directly:\n\n```\nnpx -y @prompt-ot/mcp\n```\n\nBrowser-based clients can use PromptOT’s hosted MCP endpoint with OAuth:\n\n```\nhttps://mcp.promptot.com/mcp\n```\n\nComplete installation instructions are available in the [PromptOT MCP documentation](https://www.promptot.com/docs/mcp).\n\nGiving an AI assistant access to production prompt management requires clear boundaries.\n\nWe designed PromptOT MCP around several safety principles.\n\nMCP keys can be limited to the permissions they need.\n\nA read-only assistant does not need permission to modify or publish prompts.\n\nNew MCP keys use development-oriented defaults. Publishing access must be granted intentionally.\n\nDelete operations are marked as destructive and require explicit confirmation.\n\nMCP-originated changes include client information, allowing teams to understand where each mutation came from.\n\nAn MCP key can be revoked from PromptOT. Future requests using that key will be rejected.\n\nEvery MCP tool has a defined input schema.\n\nThe assistant cannot send arbitrary database operations or execute unrestricted code through the PromptOT MCP server.\n\nThese boundaries are important because MCP should make development faster without removing control from the user.\n\nPromptOT supports two complementary paths.\n\nA production application retrieves its compiled prompt at runtime:\n\n``` js\nimport { PromptOT } from \"@prompt-ot/sdk\";\n\nconst promptot = new PromptOT({\n  apiKey: process.env.PROMPTOT_API_KEY,\n});\n\nconst { prompt, version } = await promptot.prompts.get(\n  \"support-agent\",\n  {\n    variables: {\n      user_name: \"Alex\",\n      plan: \"Pro\",\n      tone: \"friendly\"\n    }\n  }\n);\n```\n\nThe application can send the resulting prompt to its selected model provider:\n\n``` js\nconst response = await openai.chat.completions.create({\n  model: \"gpt-4o\",\n  messages: [\n    {\n      role: \"system\",\n      content: prompt\n    },\n    {\n      role: \"user\",\n      content: userMessage\n    }\n  ]\n});\n```\n\nPromptOT returns a compiled string, so the application is not locked into one model provider.\n\nClaude, Cursor or Codex uses MCP to help developers manage prompt resources through natural language.\n\nThe API serves the application.\n\nMCP serves the developer’s AI tools.\n\nBoth operate on the same PromptOT source of truth.\n\nBuilding an MCP server is not only about wrapping REST endpoints.\n\nA useful MCP implementation also needs careful tool and permission design.\n\nA small, predictable tool is easier for an AI assistant to use correctly than one large operation with many unrelated behaviours.\n\nFor example, PromptOT provides separate tools for:\n\nThis makes the assistant’s intended action easier to understand and audit.\n\nThe model relies on tool descriptions to decide which operation it should call.\n\nTool descriptions are therefore part of the product interface, not merely internal documentation.\n\nRead operations, mutations and destructive operations should be clearly distinguished.\n\nAn AI client should understand whether a tool only retrieves information or permanently changes a resource.\n\nCompiled prompts can become large.\n\nAn MCP server must return enough context for the assistant to complete its task without flooding the model context window with unnecessary data.\n\nPromptOT supports prompt truncation controls when retrieving large compiled prompts.\n\nLocal clients such as Claude Desktop, Cursor and Codex work well with a standard input/output MCP transport.\n\nBrowser-based clients require a hosted HTTP transport with authentication.\n\nBoth transports should expose the same tool catalog so the assistant receives consistent PromptOT capabilities regardless of where it is running.\n\nPromptOT MCP creates a workflow where the dashboard, application and AI development tools operate on the same prompt resources.\n\nA prompt edited from Claude appears in PromptOT’s history.\n\nA test case created from Cursor becomes available to the team.\n\nA version prepared from Codex can still be reviewed before it is published.\n\nThe goal is not to move prompt management into one specific AI client.\n\nThe goal is to maintain one source of truth while allowing developers to work from the tools they already use.\n\nAfter connecting PromptOT, start with something simple:\n\n```\nList my PromptOT prompts and explain the purpose of each one.\n```\n\nThen inspect a specific prompt:\n\n```\nOpen the customer-support prompt and summarize its role,\ninstructions, variables and guardrails.\n```\n\nCreate a safe change:\n\n```\nAdd a guardrail that prevents the assistant from requesting\npasswords. Do not change any other blocks.\n```\n\nSave a version:\n\n```\nSave the current prompt as a new draft version with the\nchangelog note “Added password safety guardrail.”\n```\n\nFinally, create test cases:\n\n```\nCheck the existing test cases for the customer-support prompt.\n\nCreate only the missing cases for:\n\n1. A normal product question\n2. A billing escalation\n3. A prompt-injection attempt\n```\n\nThese requests allow an AI assistant to work with structured prompt resources instead of copying untracked prompt text between conversations.\n\nIf you already use Claude, Cursor, Codex or another MCP-compatible client, you can connect it to PromptOT and start managing prompts from your existing workflow.\n\nExplore PromptOT:\n\nRead the MCP documentation:\n\n[https://www.promptot.com/docs/mcp](https://www.promptot.com/docs/mcp)\n\nDownload the Claude Desktop extension:\n\n[https://www.promptot.com/downloads/promptot.mcpb](https://www.promptot.com/downloads/promptot.mcpb)\n\nProduction prompts should not be scattered across source files, documents and AI conversations.\n\nWith PromptOT MCP, your dashboard, application and AI development tools can finally work from the same structured and versioned prompt workspace.", "url": "https://wpnews.pro/news/we-added-mcp-to-promptot-manage-production-prompts-from-claude-cursor-and-codex", "canonical_source": "https://dev.to/promptot/-we-added-mcp-to-promptot-manage-production-prompts-from-claude-cursor-and-codex-134d", "published_at": "2026-08-03 08:46:34+00:00", "updated_at": "2026-08-03 09:14:19.696942+00:00", "lang": "en", "topics": ["developer-tools", "ai-tools", "ai-infrastructure", "large-language-models"], "entities": ["PromptOT", "Claude", "Cursor", "Codex", "Model Context Protocol"], "alternates": {"html": "https://wpnews.pro/news/we-added-mcp-to-promptot-manage-production-prompts-from-claude-cursor-and-codex", "markdown": "https://wpnews.pro/news/we-added-mcp-to-promptot-manage-production-prompts-from-claude-cursor-and-codex.md", "text": "https://wpnews.pro/news/we-added-mcp-to-promptot-manage-production-prompts-from-claude-cursor-and-codex.txt", "jsonld": "https://wpnews.pro/news/we-added-mcp-to-promptot-manage-production-prompts-from-claude-cursor-and-codex.jsonld"}}