Do I need a server ID?

Not for https://mcp.flowra.dev/mcp. The first sign-in creates a default META server. A pinned tool set uses the same host: https://mcp.flowra.dev/mcp/{serverId}.

Is this OAuth like some other MCP hosts?

Yes for Cursor, Claude, and other clients that speak MCP OAuth: add the URL with no headers, sign in on Flowra, and pick a project. An API key still works for CI. Toolkit OAuth (Gmail, Slack) is a separate step via FLOWRA_MANAGE_CONNECTIONS.

My client is not in the list.

Any MCP client that speaks streamable HTTP can add https://mcp.flowra.dev/mcp. Omit headers for OAuth, or send x-api-key for CI.

One MCP URL for the project

https://mcp.flowra.dev/mcp is streamable HTTP. First request creates a default META server for the project. A pinned server is the same host: https://mcp.flowra.dev/mcp/{serverId}.

Start for free

Read the docs

Any MCP client

Paste this into your agent

It fetches this page, follows the setup for the client it is running in, and wires itself up.

or add the endpoint yourself

Sign in with Flowra, or a project API key
50,000+ tools over one endpoint
1,000+ apps on one connection

Pick your client

Connect your agent to Flowra tools

Each page has the exact command or config for that client, then how to check the tools are live.

How it works

From one URL to live tools

MCP server URL

mcp.flowra.dev/mcp

✓ streamable HTTP

Register the URL

Add https://mcp.flowra.dev/mcp as a streamable HTTP server in the client. Exact JSON is on each client page.

Flowra sign-in

Google · GitHub · email

✓ project connected

Sign in with Flowra

Leave headers empty. The client opens Google, GitHub, or email, then you pick a project. An API key is only for CI.

your agent · flowra

FLOWRA_DISCOVER_TOOLS
FLOWRA_GET_TOOL_SCHEMAS
✓ ready to run

Discover, then run

FLOWRA_DISCOVER_TOOLS → FLOWRA_GET_TOOL_SCHEMAS → FLOWRA_MANAGE_CONNECTIONS if an app is disconnected → FLOWRA_MULTI_EXECUTE_TOOL.

Meta tools

Six meta tools. The catalog on demand.

The client does not load the whole catalog. Four core tools discover, load schemas, connect, and run. Two sandbox tools script bulk work over saved results.

Discover
FLOWRA_DISCOVER_TOOLS
Search the catalog (Gmail, Slack, GitHub, …) without loading every tool into context.
Schemas
FLOWRA_GET_TOOL_SCHEMAS
Fetch arguments before you call. Do not invent fields.
Connect
FLOWRA_MANAGE_CONNECTIONS
Start OAuth for a toolkit and hand the user the link. App OAuth is not MCP login.
Run
FLOWRA_MULTI_EXECUTE_TOOL
Execute one or more tools. Risky sends can still wait on a Flowra approval gate.
Script
FLOWRA_JS_REMOTE_WORKBENCH
JavaScript in a remote sandbox that keeps state. Loop app tools over many items or parse a large saved result.
Shell
FLOWRA_REMOTE_BASH
One shell command over the same sandbox files: jq, grep, unzip. No app tools and no state between calls.

Agent skill

Teach the agent the workflow too.

The Flowra skill tells Cursor, Claude Code, Codex, and other skills.sh clients when to discover, connect, build, and debug. Install it next to the MCP server.

For AI agents

This page is the instruction set. Point your agent at it.

Tell your agent to set up Flowra from this URL. The same text you are reading is what it follows.

flowra.dev/mcp

## For AI agents reading this page

You are reading https://flowra.dev/mcp. Flowra is an AI agent and workflow platform: one sentence, connected apps, an approval gate before risky sends, a run ledger.

1. Register the MCP server in the client you are running in. URL:
   https://mcp.flowra.dev/mcp
   No headers. The client opens Flowra sign-in (Google, GitHub, or email); the user picks a project. For CI, x-api-key and x-username: project_default_user still work. Do not invent a key.
2. Verify before you continue. List MCP tools. You should see FLOWRA_DISCOVER_TOOLS, FLOWRA_GET_TOOL_SCHEMAS, FLOWRA_MANAGE_CONNECTIONS, FLOWRA_MULTI_EXECUTE_TOOL, FLOWRA_JS_REMOTE_WORKBENCH, FLOWRA_REMOTE_BASH. If the server is not connected, stop and say so.
3. Find the job with FLOWRA_DISCOVER_TOOLS, load schemas with FLOWRA_GET_TOOL_SCHEMAS, connect with FLOWRA_MANAGE_CONNECTIONS if needed, then FLOWRA_MULTI_EXECUTE_TOOL.
4. Tell the user which apps the connection can reach and what you can do next.

Reference: https://mcp.flowra.dev/mcp · https://docs.flowra.dev/guides/mcp · https://flowra.dev/mcp

Frequently asked questions

Give your agent the tools

One URL, sign in, then discover. Start free.

Start for free

Read the docs

50,000+ tools
One endpoint
Flowra sign-in

All clients

One MCP URL for the project

https://mcp.flowra.dev/mcp is streamable HTTP. First request creates a default META server for the project. A pinned server is the same host: https://mcp.flowra.dev/mcp/{serverId}.

Any MCP client

Paste this into your agent

It fetches this page, follows the setup for the client it is running in, and wires itself up.

or add the endpoint yourself

Sign in with Flowra, or a project API key
50,000+ tools over one endpoint
1,000+ apps on one connection

From one URL to live tools

MCP server URL

mcp.flowra.dev/mcp

✓ streamable HTTP

Register the URL

Add https://mcp.flowra.dev/mcp as a streamable HTTP server in the client. Exact JSON is on each client page.

Flowra sign-in

Google · GitHub · email

✓ project connected

Sign in with Flowra

Leave headers empty. The client opens Google, GitHub, or email, then you pick a project. An API key is only for CI.

your agent · flowra

FLOWRA_DISCOVER_TOOLS
FLOWRA_GET_TOOL_SCHEMAS
✓ ready to run

Discover, then run

FLOWRA_DISCOVER_TOOLS → FLOWRA_GET_TOOL_SCHEMAS → FLOWRA_MANAGE_CONNECTIONS if an app is disconnected → FLOWRA_MULTI_EXECUTE_TOOL.

Six meta tools. The catalog on demand.

The client does not load the whole catalog. Four core tools discover, load schemas, connect, and run. Two sandbox tools script bulk work over saved results.

Discover

FLOWRA_DISCOVER_TOOLS

Search the catalog (Gmail, Slack, GitHub, …) without loading every tool into context.

Schemas

FLOWRA_GET_TOOL_SCHEMAS

Fetch arguments before you call. Do not invent fields.

Connect

FLOWRA_MANAGE_CONNECTIONS

Start OAuth for a toolkit and hand the user the link. App OAuth is not MCP login.

Run

FLOWRA_MULTI_EXECUTE_TOOL

Execute one or more tools. Risky sends can still wait on a Flowra approval gate.

Script

FLOWRA_JS_REMOTE_WORKBENCH

JavaScript in a remote sandbox that keeps state. Loop app tools over many items or parse a large saved result.

Shell

FLOWRA_REMOTE_BASH

One shell command over the same sandbox files: jq, grep, unzip. No app tools and no state between calls.

This page is the instruction set. Point your agent at it.

Tell your agent to set up Flowra from this URL. The same text you are reading is what it follows.

flowra.dev/mcp

## For AI agents reading this page

You are reading https://flowra.dev/mcp. Flowra is an AI agent and workflow platform: one sentence, connected apps, an approval gate before risky sends, a run ledger.

1. Register the MCP server in the client you are running in. URL:
   https://mcp.flowra.dev/mcp
   No headers. The client opens Flowra sign-in (Google, GitHub, or email); the user picks a project. For CI, x-api-key and x-username: project_default_user still work. Do not invent a key.
2. Verify before you continue. List MCP tools. You should see FLOWRA_DISCOVER_TOOLS, FLOWRA_GET_TOOL_SCHEMAS, FLOWRA_MANAGE_CONNECTIONS, FLOWRA_MULTI_EXECUTE_TOOL, FLOWRA_JS_REMOTE_WORKBENCH, FLOWRA_REMOTE_BASH. If the server is not connected, stop and say so.
3. Find the job with FLOWRA_DISCOVER_TOOLS, load schemas with FLOWRA_GET_TOOL_SCHEMAS, connect with FLOWRA_MANAGE_CONNECTIONS if needed, then FLOWRA_MULTI_EXECUTE_TOOL.
4. Tell the user which apps the connection can reach and what you can do next.

Reference: https://mcp.flowra.dev/mcp · https://docs.flowra.dev/guides/mcp · https://flowra.dev/mcp