v0 MCP × Kiro CLI | Vercel x AWS | agent 2 agent design fun
Vercel's v0 MCP server can be used with Kiro CLI agents — even though Kiro is not on Vercel's official supported-clients list or documentation
AWS kiro-cli writes code and can call out to external tools. Vercel hosts websites & makes v0 "text-to-app" builder. V0 allows description of a web page in English to be generated into a working app with a real time live preview. Model Context Protocol provides a standard allowing Kiro agents to use v0 as a tool that talks back
The "traditional" way people use the v0 website v0.dev is to type in a browser, watch the preview update. But programmatic access is available via https://api.v0.dev
v0's server speaks Bearer-token auth, granting access via an API key generated in the v0 site. This enables mcp-remote bridges of any standard input/output MCP client
kiro (left) is sending chats via MCP and v0 (right) is responding and acting
mcp.v0.dev
The server —Vercel's official MCP endpoint— enables five tools for kiro-cli: create_chat, send_message, get_chat, list_chats, find_chats. Auth is a bearer token titled V0_API_KEY
This is distinct from Vercel MCP (mcp.vercel.com), which is OAuth-only and allowlisted-clients-only. Vercel MCP (mcp.vercel.com) only admits pre-approved clients: Claude, Cursor, Copilot, Goose, Windsurf, Gemini CLI… Kiro is not on that list. But v0 MCP (mcp.v0.dev) using bearer-token auth instead of OAuth — unlocking work with any stdio MCP client, Kiro included.
SSM → mcp-remote
Rather than as a static environmental variable in our shell profile, we stored our v0 API key in AWS SSM organized with /mcp/v0-api-key (SecureString, profile, region to make calls with AWS SSM). A launcher fetches it at runtime via get-parameter --with-decryption, then runs npx mcp-remote with a --header flag injecting the Bearer token. mcp-remote bridges stdio ⇆ Streamable HTTP
Rather than as a static environmental variable in our shell profile, we stored our v0 API key in AWS SSM organized with /mcp/v0-api-key (SecureString, profile, region to make calls with AWS SSM). A launcher fetches it at runtime via get-parameter --with-decryption, then runs npx mcp-remote with a --header flag injecting the Bearer token. mcp-remote bridges stdio ⇆ Streamable HTTP
kiro's mcp.json (i.e. ~/.kiro/settings/mcp.json)
The kiro-cli mcp.json can be set up to register v0 as a bash -c command pointing at a launcher
"v0": { "command": "bash", "args": ["-c", "/home/username/mcp-launchers/run.sh v0.sh"], "description": "v0 MCP Server — Vercel AI UI code generation. Create/manage v0 chats, generate React/UI components. Requires V0_API_KEY env var." }The launcher —A small bash script fetches the key from SSM into a local variable and execs mcp-remote against https://mcp.v0.dev with the Authorization header
v0 MCP —create_chat starts a new v0 generation chat from a prompt; send_message continues a chat (iterate on components); get_chat retrieves a chat + its generated code; list_chats browses existing chats; find_chats searches chats by content
We used v0 MCP to generate React/Next.js components that inherit our material design based system. Kiro seeds a chat with our token context + UX card specs, iterates on the component output, and retrieves the generated code —without leaving the agent session
back at https://v0.app/ we can see v0 responding to kiro as well as a preview
v0 chat
v0 Tokenomics
Every create chat and send message call can name which v0 model does the work, via a modelConfiguration object. v0-mini is lightning-fast, near-frontier; cheapest. v0-auto lets v0 pick the tier to fit the task. v0-pro provides balanced speed and intelligence; a good default for most work. v0-max Top capability — and most expensive excepting... v0-max-fast provides high capability, tuned for speed
Two switches in the same config add cost on top of the tier: thinking (extended reasoning) and imageGenerations. Both default to false. These friendly names sit on top of v0's underlying "composite model family" — models that blend retrieval, a frontier LLM's reasoning, and a custom error-fixing pass
The cost model is per-token, not per-call. v0 bills input and output tokens separately, with discounted rates for cache hits. The smallest tier (v0 Mini) is roughly $1 per million input tokens and $5 per million output tokens; higher tiers cost more per token. The practical consequence: spend scales with how big the prompt is × how much code is generated × how expensive the tier is. A large prompt that regenerates many files on the top tier is the fastest way to burn credits
While building this site we sent our v0 messages on v0-max-fast — including one very large bundled message that rebuilt several pages at once. Top tier multiplied by a huge generation spent $100 in one morning before the morning cup of joe was empty. The fix was not "use the API differently" — it was "pick the right model." A UI-polish pass does not need the top reasoning tier; v0-auto or v0-pro would have produced comparable output for a fraction of the spend
Smaller prompts cost less. One focused change beats one giant everything-at-once message — and it is easier to review.
Start a chat from existing files, rather than generating from scratch, is fast and spends no tokens. We spent weeks setting up references we ingested as a style system before ever prompting v0
BONUS: AWS MCP
Agents building project Ðekawɔwɔ for the #H0Hackathon need reach into two AWS accounts — authenticated, least-privilege, no embedded credentials. This allowed Kiro to query infrastructure, run CDK deploys, sync previews, and pull live docs without leaving the session
The Managed AWS MCP Server (GA 2026-05-06), is reached via mcp-proxy-for-aws, exposing 10 tools over endpoint https://aws-mcp.us-east-1.api.aws/mcp
The auth model leverages our IAM Identity Center short-lived sign-ons via SigV4 from local ~/.aws SSO profiles, bridged to the OAuth-only MCP endpoint by mcp-proxy-for-aws. No static API keys.
Kiro agents then switch profiles per-call across two accounts, giving direct, authenticated, least-privilege access to both AWS accounts without embedding credentials. The agents use it to deploy CDK stacks, check infrastructure health, sync S3 previews, manage Cognito users, query Aurora DSQL / DynamoDB, and pull live AWS documentation for the specific services Ðekawɔwɔ uses (Route53, Location Service, Bedrock Nova)
This content was created following entry to the H0: Hack the Zero Stack with Vercel v0 and AWS Databases hackathon. "Front-end in minutes. back-end designed to scale." #H0Hackathon
mcp provided Ðekawɔwɔ with agentic access to aws apis & current documentation. Thanks for reading - ^.^
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article