Operate your tenant
Query entity records, list FSM jobs by state, fire transitions, and read the append-only event log and notifications — so an assistant can triage what's happening and move work forward.
Agent-ready
Let AI operate your business — safely.
Connect Claude — or any Model Context Protocol host — to your tenant. It can read your data and drive your workflows, but every change is guarded: dry-run previews before anything commits, your FSM guards stay in charge, and nothing destructive happens on its own.
Your assistant, on your terms
FastYoke speaks MCP — the emerging open standard for connecting AI assistants to real systems — over the same token-scoped API the SDK uses. Bring your own assistant; your data never leaves your tenant; and the platform's guards, not the model, decide what's allowed.
Point an assistant at a scoped API token and it can help run the operation you already built — reading what's there and proposing the next move — without you handing over the keys.
What an assistant can do
Operate and build — through one server
The same server exposes two surfaces: running your live operation, and shaping the apps behind it.
Build on the platform
Introspect entity schemas and FSM definitions, validate a candidate schema, and author a new schema version — assistant-assisted app design, over the same open contract.
Safe by construction
An assistant that can act — without the risk
"Agent-ready" only matters if it's safe. Control is a property of the architecture here, not a prompt you hope holds.
Dry-run first
Every write tool takes a dry-run flag. The assistant can preview exactly what a change would do — the validation result and the resulting state — before anything commits.
Your guards stay authoritative
A transition is evaluated by your server-side FSM guards, not the model. An assistant cannot fire a state change your workflow doesn't allow — the engine decides.
No autonomous destruction
The administrative cancel-override — forcing a job to a terminal state past its guards — is off-limits to assistants. Record deletion is opt-in and requires explicit confirmation.
Read-only mode
Flip one setting and no write tool is registered at all. The assistant can observe and reason about your operation, but it cannot change a thing.
Token-scoped, tenant-bound
A narrow API token is the whole boundary — scope it down and that's exactly what the assistant can touch. Nothing reaches beyond your tenant, and the token is never logged or echoed back.
How to connect
Add one server to your assistant
The server runs locally over stdio and is launched by your MCP host — Claude Desktop, Claude Code, or any other. Point it at your tenant with a scoped token:
{
"mcpServers": {
"fastyoke": {
"command": "npx",
"args": ["-y", "@fastyoke/mcp"],
"env": {
"FASTYOKE_BASE_URL": "https://app.fastyoke.io",
"FASTYOKE_PAT": "fy_pat_…",
"FASTYOKE_TENANT_ID": "ten_…"
}
}
}
}
Mint the token from your tenant and scope it to what the assistant should reach. The full setup — every tool, every environment variable, and the guarded-write model — is in the developer docs.
Available now
What's here — and what's next
The MCP server is available now. Here today: the operate and build tools above, guarded writes with dry-run, and read-only mode. What's next — form submission, guided app scaffolding, MCP prompts, and a FastYoke-hosted remote endpoint so there's nothing to run locally.
If there's a workflow you'd want an assistant to handle, tell us — it helps us sequence what ships next.
Make your operation agent-ready
Connect an assistant to the operation you already run — safely, on your terms, today.