Build Intelligence · inbound MCP
Build Rylvo from the AI editor you already use.
Connect a compatible Streamable HTTP MCP client to Rylvo. Discover 129 typed tools for bots, prompts, guardrails, skills, knowledge, tests, channels, workflows, and operations—inside the workspace you authorize.
129
live MCP tools
5
verified client setups
3
OAuth scopes
24h
access-token lifetime
How it works
From editor prompt to controlled action.
Build Intelligence is Rylvo's inbound MCP surface: your external AI client discovers typed workspace tools and calls them through an authorized organization context.
Configure /mcp
Add Rylvo's remote Streamable HTTP endpoint to a compatible client.
Authorize in browser
Sign in, choose the workspace, review requested scopes, and approve access.
Discover typed tools
The client lists 129 live tools and their input schemas from the server.
Call with controls
Scope, org, plan, quota, and input checks run before the underlying action.
https://rylvo.com/mcp plus OAuth discovery.Client setup
Use the configuration your client actually expects.
These examples match the current configuration shape for each client. Pick one, connect, then approve Rylvo in the browser.
Claude Code
Terminal, then /mcp inside Claude Code
claude mcp add --transport http rylvo https://rylvo.com/mcp # In Claude Code, run /mcp and choose Authenticate
The explicit HTTP transport flag avoids the command being interpreted as a local stdio server.
129 live tools
Build, connect, test, and operate—without hiding the primitives.
The server exposes typed tools across the workspace. Your client can inspect each schema before it calls anything, and Rylvo returns structured success or error results.
Build the agent
Create, inspect, clone, update, and remove bots plus their runtime settings.
Bots · templates · settings · cloning
Shape behavior
Manage versioned prompts, guardrails, skills, optimization settings, and edge cases.
Prompts · guardrails · skills · evolution
Ground it in data
Connect knowledge, manage sources and datasets, and inspect workspace resources.
Knowledge · sources · datasets
Wire integrations
Register external MCP servers, configure connectors, channels, API keys, and webhooks.
MCP · connectors · channels · keys
Test before release
Create cases, run suites, read results, and use production edge cases as test inputs.
Test cases · suite runs · results
Operate the workspace
Manage flows, workflows, agent groups, automations, broadcasts, and oversight resources.
Flows · automation · Mission Control
Useful composed actions are included too
scaffold_bot_from_description creates a bot, prompt, and starter guardrail; clone_bot can copy linked resources; attach_mcp_server registers a server and links it to bots.
Security model
A real credential boundary, not a magic tunnel.
OAuth makes the connection manageable and inspectable. It does not make a leaked credential harmless, so the page now states the actual boundary and lifetime plainly.
Org and audience bound
The signed access token names one organization, client, scope set, and MCP resource audience.
Three scopes
rylvo:read, rylvo:write, and rylvo:admin keep discovery, mutation, and destructive/admin work separate.
Rotating credentials
Access tokens last 24 hours. Refresh tokens rotate, use a 30-day sliding window, and have a 365-day family ceiling.
PKCE and revocation
Public clients use PKCE S256 in production. Workspace admins can revoke a connected OAuth client.
Auditable tool calls
Tool handlers emit success/error, duration, and redacted argument summaries into Rylvo's audit path.
Rate limited
The public transport is limited per IP; each tool is also limited per organization and tool name.
rylvo:readInspect workspace resources
rylvo:writeCreate and update resources
rylvo:adminDelete and perform admin operations
Direct API integration
OpenAPI is a reference, not an MCP synonym.
Rylvo also publishes a broader OpenAPI 3.1 engine specification. Use it to select REST operations for your application, then apply the authentication required by those routes.
OpenAPI 3.1 specification
A machine-readable snapshot of the broader engine API for developer integration and code generation.
https://rylvo.com/openapi.json
Choose the right surface
Four paths. Four different jobs.
Use this map when deciding whether the intelligence lives in an external editor, inside Rylvo, inside a deployed bot, or in your own service.
AI editor → Rylvo
Build Intelligence
Use 129 org-scoped MCP tools from a compatible external editor or client.
Manage connectionsRylvo UI → Rylvo
Workspace Architect
Plan and apply workspace changes in the built-in chat, using the same MCP-native tool family.
Explore ArchitectRylvo bot → external tools
MCP Hub
Connect deployed bots to third-party MCP servers and choose which external tools they can call.
Explore MCP HubYour service → Rylvo API
OpenAPI reference
Use the broader REST contract for direct application integration and selected server-to-server calls.
Open the specFAQ
What to know before you connect
The operational details that matter when an AI client can change a real workspace.
Does Build Intelligence bypass plan limits?
No. MCP is another controlled entry point to the same product capabilities. A tool can be rejected by OAuth scope, workspace role, feature availability, quota, validation, or the underlying service.
Can a token move between organizations?
No. The access token is bound to the organization selected during consent and requests execute in that organization. A stolen token could still act inside that organization until it expires or is revoked, so treat it as a credential.
Are destructive calls automatically paused for approval?
Rylvo requires the admin scope for delete and other admin tools, but the MCP endpoint does not add a separate approval pause to every call. Use the client's tool-confirmation controls and grant only the scopes you need.
Is /openapi.json the same as the MCP tool catalog?
No. It is a broader OpenAPI 3.1 engine reference, not a one-to-one export of the curated MCP tools. Review the operations you expose and supply each route's required authentication.
Which clients have setup instructions here?
Claude Code, Cursor, Windsurf, Zed, and Codex CLI. Other standards-compliant Streamable HTTP MCP clients may work, but their configuration and OAuth support should be checked with that client.
How do I disconnect a client?
Open Dashboard → Build Intelligence → Connected clients. An admin or owner can revoke the client, which invalidates its refresh-token family; issued access tokens are also checked against revocation.
Make your editor a Rylvo workspace client.
Connect through OAuth, grant the minimum scopes you need, and manage the client from one dashboard.