Rylvo, Inc.

Build Intelligence

Your AI editor, connected to Rylvo

Q: Which AI editors are supported?

Claude Code, Cursor, Windsurf, and Zed are supported as MCP-native clients with OAuth handled automatically. Any other function-calling model (OpenAI Responses, Google Gemini, Moonshot Kimi) can connect via the OpenAPI 3.1 spec.

Q: How is authentication handled?

OAuth 2.1 with PKCE (S256). MCP clients trigger the browser flow automatically on first install. Access tokens are RS256 JWTs with a one-hour TTL, bound to a single org. Refresh tokens rotate on every exchange.

Q: What can my AI actually do through the MCP connection?

30 resource tools for bots, prompts, guardrails, MCP servers, workflows, skills, and knowledge-base connections — covering list, get, create, update, and delete operations. Plus 3 macro tools that compose multiple steps: scaffold_bot_from_description, clone_bot, and attach_mcp_server.

Q: Is every call scoped to my organization?

Yes. Every token carries an org_id claim, and every tool call is authorized against that claim before it executes. There is no cross-org access even if a token leaks. A 403 is returned the moment the claim and the resource org diverge.

Q: Can I connect without MCP?

Yes. Every endpoint is documented in an OpenAPI 3.1 spec served at /openapi.json. Any model that supports function calling can ingest the spec and gain typed tools — OpenAI Responses, Gemini, Kimi, and open-source tool-using LLMs are all supported.

Connect Claude Code, Cursor, Windsurf, Zed, or any function-calling AI to Rylvo via MCP or OpenAPI. Your editor gains 30 tools to create and manage bots, prompts, guardrails, skills, and more — all scoped to your org.

Get Started See MCP Hub

Works with the editors you already use

Claude Code

Cursor

Windsurf

Zed

OpenAI Codex

OpenAI Responses

Google Gemini

Natural-language control plane

Your editor becomes a Rylvo operator. Create bots, tune prompts, configure guardrails, and wire up MCP servers by describing what you want — no context-switching to the dashboard.

OAuth 2.1 + PKCE, by default

MCP-native clients trigger the browser flow automatically. Access tokens are short-lived RS256 JWTs bound to a single org. Refresh tokens rotate on every exchange.

Org-scoped on every call

Every tool invocation re-verifies the token's org_id against the target resource. A leaked token can never reach another tenant — the 403 lands before the query runs.

How it works

One command to 30 tools

From install to first natural-language build in under a minute. No tokens to paste, no API keys to rotate.

Run install

one command in your editor

OAuth opens

browser redirects to Rylvo

Token issued

RS256 JWT · 1 hr TTL

30 tools ready

scoped to your org

Build in NL

bots, prompts, skills…

Clients

MCP-native install, one command

OAuth is handled automatically. Pick your editor, copy the line, and your browser does the rest.

Claude Code

Recommended

One command — browser opens for OAuth automatically.

shell

claude mcp add rylvo https://api.rylvo.com/mcp

Capabilities

30 tools your AI can call, scoped to your org

Read, write, and delete across every Rylvo primitive — plus three macro tools that compose common multi-step operations.

Resource	Read	Write	Delete
Bots	bots.list, bots.get	bots.create, bots.update	bots.delete
Prompts	prompts.list, prompts.get	prompts.create, prompts.update	prompts.delete
Guardrails	guardrails.list, guardrails.get	guardrails.create, guardrails.update	guardrails.delete
MCP Servers	mcp_servers.list	mcp_servers.register	mcp_servers.delete
Workflows	workflows.list	workflows.create	workflows.delete
Skills	skills.list	skills.create	skills.delete
Knowledge Base	kb.list_connections	kb.create_connection, kb.add_source	—

scaffold_bot_from_description

One call: bot + system prompt + starter guardrail from a natural-language brief.

clone_bot

Duplicates a bot; deep_copy_resources=true also copies linked prompts and guardrails.

attach_mcp_server

Registers an external MCP server, writes its tool list, and links it to bots in one call.

OpenAPI 3.1

Not using MCP? Use function-calling

OpenAI Responses, Google Gemini, Moonshot Kimi, and any tool-using LLM can ingest the spec at /openapi.json and gain typed tools for every Rylvo endpoint.

Same auth, same org-scoping, same 30-plus capabilities — just served in the shape your model already speaks.

Spec URL

python

import httpx, openai
spec = httpx.get("https://api.rylvo.com/openapi.json").json()
# map operations → function tools, pass to
# client.responses.create(tools=[...])

Auth

Manual OAuth 2.1 + PKCE, for CLI agents

Build your own MCP or API client? Here's the full flow — PKCE S256 is required in production.

1

Register your client (one-time)

shell

curl -X POST https://api.rylvo.com/oauth/register \
  -H 'Content-Type: application/json' \
  -d '{"client_name":"my-cli","redirect_uris":["http://localhost:8765/callback"],"token_endpoint_auth_method":"none"}'

Response includes client_id — no secret needed for PKCE public clients.

2

Open authorize URL in the browser

shell

https://api.rylvo.com/oauth/authorize
  ?response_type=code
  &client_id=<your-client-id>
  &redirect_uri=http://localhost:8765/callback
  &scope=rylvo:read+rylvo:write
  &code_challenge=<S256 of verifier>
  &code_challenge_method=S256

After consent, the user is redirected to your redirect_uri with ?code=…

3

Exchange code for tokens

shell

curl -X POST https://api.rylvo.com/oauth/token \
  -d grant_type=authorization_code \
  -d client_id=<id> \
  -d code=<code> \
  -d redirect_uri=http://localhost:8765/callback \
  -d code_verifier=<verifier>

Returns access_token (1 hr) + refresh_token. Use as Authorization: Bearer {token}.

4

Refresh (tokens rotate)

shell

curl -X POST https://api.rylvo.com/oauth/token \
  -d grant_type=refresh_token \
  -d client_id=<id> \
  -d refresh_token=<rfr-token> \
  -d org_id=<your-org-id>

Refresh tokens rotate on every exchange — always store the new one.

Security

Short-lived, org-scoped, auditable

Every access token is a narrowly-scoped JWT. Every call is logged. Every refresh rotates.

RS256 JWT

Asymmetric signing. Public JWKS for verification without sharing secrets.

1-hour TTL

Access tokens expire quickly. Leaked tokens become useless within the hour.

PKCE S256

Public clients — no embedded secrets. Plain challenges rejected in production.

Org-scoped

Every tool call re-verifies the token's org_id against the resource being touched.

FAQ

Questions about Build Intelligence

Everything you need to know before your AI editor makes its first Rylvo call.

Which AI editors are supported?

Claude Code, Cursor, Windsurf, Zed, and Codex CLI are first-class MCP clients with OAuth handled automatically. Any other function-calling model — OpenAI Responses, Google Gemini, Moonshot Kimi, or open-source tool-using LLMs — can connect via the OpenAPI 3.1 spec.

How is authentication handled?

OAuth 2.1 with PKCE (S256 only in production). MCP-native clients open the browser flow automatically on first install. Access tokens are RS256 JWTs with a one-hour TTL, bound to exactly one org. Refresh tokens rotate on every exchange.

Is every call scoped to my org?

Yes. Every JWT carries an org_id claim that is re-verified on each request against the resource being touched. A token that leaks cannot reach another tenant — the middleware returns 403 before the handler even runs.

Can I connect without MCP?

Yes. The OpenAPI 3.1 spec at /openapi.json documents every endpoint. Point your model's tool loader at it and you get typed function-calling across the entire Rylvo API.

What do the 3 macro tools do?

They compose frequent multi-step operations. scaffold_bot_from_description creates a bot plus system prompt plus starter guardrail from a brief. clone_bot duplicates a bot and optionally its linked resources. attach_mcp_server registers an external server, writes its tool list, and links it to bots — all in one call.

Ready to hand your editor the keys to Rylvo?

Install once. OAuth handles itself. Your AI gets 30 tools scoped to your org, with every call signed, auditable, and short-lived.

Get Started Read the Docs