Rynko
Rynko
Sign InGet Started
MCP Integration

Rynko MCP: Flow Gates + Document Generation

Connect Claude Desktop, Cursor, Windsurf, or Microsoft 365 Copilot directly to Rynko via MCP. Submit runs to Flow gates, validate AI outputs, generate documents, and manage environments through natural conversation.

Get StartedConnect to Rynko
Flow MCPhttps://api.rynko.dev/api/flow/mcp
Render MCPhttps://api.rynko.dev/api/mcp-documents
npm|GitHub

Building multi-agent pipelines with LangGraph, CrewAI, or AutoGen?

Each Flow gate is auto-exposed as an MCP tool — dynamic discovery, human-in-the-loop, circuit breaker included.

Full integration guide

How It Works

Two separate MCP endpoints — one for Flow gate validation, one for Render document generation. Connect either or both from your AI tool.

Claude
Cursor
Windsurf
VS Code
Zed
M365 Copilot
MCP Protocol (Streamable HTTP)
Flow Gates
api/flow/mcp
submit_run · get_run_result
list_gates · get_gate
APPROVED
REJECTED
Render Engine
api/mcp-documents
generate_document
list_templates · create_template
PDF
Excel
api.rynko.dev/api/flow/mcp
api.rynko.dev/api/mcp-documents
Flow + Render Pipeline

Validate first.
Generate after.

Your AI agent validates any data through a Flow gate before taking downstream actions. If approved, it can immediately use Render to generate a document — all in one conversation.

  • Agent calls submit_run → Gate checks schema + business rules
  • APPROVED → Agent proceeds (generate doc, notify, save to CRM)
  • REJECTED → violations array explains exactly what failed
  • REVIEW_REQUIRED → routed to human approval inbox
Claude Desktop — Rynko MCP

Auto-plays · click to replay

Tested with a Real LLM

LLMs read our rejections and fix themselves.

We pointed a real LLM at our MCP server with zero error-handling prompts. It discovered the gate, submitted a bad payload, read the structured violations, and corrected every field on its own.

No retry logic was coded. No error-handling prompts were written.

vendor: """Acme Inc"
amount: -500500
currency: "JPY""USD"

Real terminal recording — click to play

Try it yourself

Paste any of these prompts into Claude, Cursor, or any MCP-connected LLM. The agent will discover your gates, submit the payload, hit validation errors, and self-correct — no retry logic needed.

Invoice with bad data

Submit an invoice to the "Invoice Validator" gate with vendor set to empty, amount as -500, and currency as JPY. Fix any issues the gate reports until it passes.

Missing required fields

Submit a run to my Flow gate with only a customer name. Leave out all other fields. When the gate rejects it, read the errors and fill in the missing fields with reasonable values.

Discover and test

List all my Flow gates, pick the first one, and submit a test payload with intentionally wrong types for each field. Then fix every violation the gate reports.

Out-of-range values

Submit an order to my gate with quantity set to 0, discount at 150%, and a delivery date in the past. Correct each field based on the validation errors.

Get Started

Two separate MCP endpoints — connect the product(s) you need. Claude Desktop uses OAuth; other clients use an API key.

Flow Gates MCP Endpoint
https://api.rynko.dev/api/flow/mcp
Via Connectors (OAuth)
Recommended
No token needed — Claude Desktop handles authentication via OAuth automatically.
1
Open Claude Desktop → Settings → Connectors
2
Click Add Custom Connector
3
Paste the server URL: https://api.rynko.dev/api/flow/mcp
4
A browser window will open — sign in with your Rynko account to authorize

Requires Claude Pro, Max, Team, or Enterprise plan.

Via Config File (PAT)
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "rynko-flow": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://api.rynko.dev/api/flow/mcp",
        "--header",
        "Authorization:${RYNKO_FLOW_KEY}"
      ],
      "env": {
        "RYNKO_FLOW_KEY": "Bearer fm_your_api_key"
      }
    }
  }
}

Uses the mcp-remote proxy to bridge the remote server. Requires Node.js and a PAT token.

Use both products together (Claude Desktop example)
Register both MCP servers in a single config. Your AI can then call Flow gate tools and Render document tools in the same conversation.
Add both as separate Connectors in Claude Desktop:

1. Settings → Connectors → Add Custom Connector
   URL: https://api.rynko.dev/api/flow/mcp

2. Settings → Connectors → Add Custom Connector
   URL: https://api.rynko.dev/api/mcp-documents

Both will authorize via OAuth automatically.
Microsoft 365 Copilot
New
Generate documents from Word, Teams, Outlook, and other M365 apps. Connect the Render MCP server to Microsoft 365 Copilot via Copilot Studio or as a declarative agent.
Copilot Studio (No-Code)

Use the MCP onboarding wizard to connect — just paste the server URL and configure API key auth.

https://api.rynko.dev/api/mcp-documents
Declarative Agent (VS Code)

Scaffold an agent with the M365 Agents Toolkit — version-controlled, full tool selection.

M365 Agents Toolkit v6.3+
Setup Guide Requires Microsoft 365 Copilot license

Start Building

Flow prompts
  • "List my Flow gates"
  • "Validate this data against the Lead Qualifier gate"
  • "Show me the last 5 rejected runs on the customer gate"
  • "What rules does the Invoice gate enforce?"
Render prompts
  • "List my templates"
  • "Create an invoice template with line items and tax"
  • "Generate a PDF from my invoice template with sample data"
  • "Generate a sales quote for Acme Corp, $12,000"

MCP Tool Reference

17 tools across both endpoints — Flow gate validation and the full document lifecycle.

Flow gate tools are auto-discovered

When a client connects to api/flow/mcp, the server calls tools/list and returns one tool per gate configured in the connected environment. Each tool's name and description come directly from the gate — so the LLM understands what each gate validates and selects the right one from context, with no hardcoded IDs or explicit discovery step in your code. Add a gate in the dashboard → it appears as a new tool in your next MCP session automatically.

Flow Gates

api/flow/mcp
list_gates

List Flow gates in current environment

get_gate

Get gate schema, rules, and configuration

submit_run

Submit data to a Flow gate for validation

get_run_result

Get the result of a Flow gate run

Environment

api/flow/mcp
list_environments

List all environments you have access to

select_environment

Switch to a different environment context

Templates

api/mcp-documents
list_templates

List all templates in current environment

get_template

Get full template details including schema

create_template

Create a new template (draft mode only)

update_template

Update an existing template

delete_template

Delete a template (drafts only)

Documents

api/mcp-documents
generate_document

Generate a PDF or Excel document

get_generation_status

Check status of a document generation job

Utilities

api/mcp-documents
list_folders

List template folders in environment

parse_data

Parse CSV/Excel data for variable extraction

get_schema_reference

Get the template schema reference

validate_template

Validate a template schema before saving

Use Cases

Real-world scenarios where AI + Rynko shines.

AI Output Validation Before Processing
You
Check if this extracted customer record is valid before we generate their onboarding documents.
AI

Submitting to the "Customer Validator" gate...

✓ APPROVED · 4/4 rules passed

All checks passed. Generating onboarding PDF now.

Invoice Generation from Conversation
You
Create an invoice for Acme Corp. 10 hours of consulting at $150/hr. Add 20% VAT.
AI

Generating invoice using your "Professional Invoice" template.

  • Subtotal: $1,500.00 · VAT (20%): $300.00 · Total: $1,800.00

PDF generated. Download link valid for 24 hours.

Validate + Generate in One Conversation
You
Validate this new supplier record against our compliance gate, then generate their onboarding contract if it passes.
AI

Running compliance gate check...

✓ APPROVED · Generating onboarding contract...

Contract PDF ready. Attached to supplier record.

Blocked by Gate — Clear Rejection Reason
You
Submit this loan application to the underwriting gate.
AI

Submitting to the "Loan Underwriting" gate...

✗ REJECTED · 2/4 rules failed
• credit_score < 650 minimum (got 580)
• debt_to_income_ratio > 0.43 limit (got 0.51)

Application blocked. The violations above explain exactly what needs to change before resubmission.

Multi-Agent Pipelines

Using LangGraph, CrewAI, or AutoGen?

Each gate you configure is exposed as an MCP tool when the session starts — auto-discovered, description-first. Your orchestrator handles routing; Flow gates handle validation.

One tool per gate, auto-discoveredHuman-in-the-loop with conditional approvalCircuit breaker via persistent SSE session
Full integration guide

Security

Enterprise-grade security by default.

Draft-Only Mode

AI agents can only create draft templates. Publishing requires human approval in the dashboard.

Short-Lived OAuth Tokens

Claude Desktop uses OAuth with short-lived access tokens. Other clients use revocable API keys managed from your dashboard.

Full Audit Trail

Every AI action is logged with timestamp and user context. View in your activity feed.

Ready to Build with AI?

Connect Claude Desktop via OAuth, or grab an API key for Cursor, Windsurf, and VS Code.

Flow:https://api.rynko.dev/api/flow/mcp
Render:https://api.rynko.dev/api/mcp-documents
Get Started FreeFull Documentation