Flowsery MCPサーバー

Flowsery agent

Let your AI agent query Flowsery, the privacy-first, cookie-free web analytics: live visitors, traffic breakdowns by 25 dimensions, goals, revenue, visitor journeys, and the bugs, broken flows and UX problems Flowsery's AI found in session recordings. 28 MCP tools, OAuth sign-in, no cookies on your site.

Connect in 30 seconds

claude mcp add --transport http flowsery https://mcp.flowsery.com/mcp

Any other MCP client needs only the URL, https://mcp.flowsery.com/mcp. The first call opens a browser to sign in with your Flowsery account. Nothing to install and no key to paste. Setup for each client is below.

What this repo is

This repo holds every Flowsery agent integration except the Claude Code and Hermes Agent plugin:

Path	What it is
skills/flowsery/	The agent skill for npx skills add. It drives the Flowsery REST API through scripts/flowsery.js with an API key.
mcp-server/	Source of the MCP server that runs at https://mcp.flowsery.com/mcp. A push to main that touches it builds a new image.
gemini-extension.json, GEMINI.md	The Gemini CLI extension. It connects to the hosted MCP server and signs in with OAuth.
Dockerfile, glama.json	Runs the MCP server over stdio, for Glama and anyone who wants it in a container.
lhm.plugin.json	The LobeHub marketplace listing for the hosted MCP server.

The Claude Code and Hermes Agent plugin lives in Flowsery/claude-plugin. It signs in with OAuth and never asks for an API key, which the Claude plugin directory and the Hermes catalog both require, so it ships its own MCP-only skill instead of the script skill here.

Pick your install

You use	Install	Sign-in
Claude Code	/plugin install flowsery --marketplace Flowsery/claude-plugin	OAuth
Hermes Agent	hermes plugins install flowsery	OAuth
Gemini CLI	gemini extensions install https://github.com/Flowsery/agent	OAuth
Claude Desktop, Cursor, ChatGPT or any MCP client	Add https://mcp.flowsery.com/mcp as a remote server	OAuth, or an API key header
Cursor, Codex, Windsurf, Claude Code or any agent with skills	npx skills add Flowsery/agent	API key

Setup by client

The server answers an unauthenticated call with a 401 that points at its OAuth metadata, so every client below opens the Flowsery sign-in on its own. One sign-in reaches every workspace you belong to.

Claude.ai and Claude Desktop

Open Settings > Connectors, add a custom connector with the URL https://mcp.flowsery.com/mcp, and sign in. A Claude Desktop build without connectors can go through mcp-remote, which handles the sign-in itself:

{
  "mcpServers": {
    "flowsery": {
      "command": "npx",
      "args": ["mcp-remote", "https://mcp.flowsery.com/mcp"]
    }
  }
}

Claude Code

The plugin adds the server and a skill that knows how to use it:

/plugin install flowsery --marketplace Flowsery/claude-plugin

Or add only the server:

claude mcp add --transport http flowsery https://mcp.flowsery.com/mcp

Run /mcp, pick flowsery and sign in.

ChatGPT

Add a custom connector in Settings > Apps & Connectors with the URL https://mcp.flowsery.com/mcp and sign in.

Cursor

~/.cursor/mcp.json:

{
  "mcpServers": {
    "flowsery": { "url": "https://mcp.flowsery.com/mcp" }
  }
}

Windsurf

~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "flowsery": { "serverUrl": "https://mcp.flowsery.com/mcp" }
  }
}

VS Code

.vscode/mcp.json:

{
  "servers": {
    "flowsery": { "type": "http", "url": "https://mcp.flowsery.com/mcp" }
  }
}

Codex

~/.codex/config.toml:

[mcp_servers.flowsery]
url = "https://mcp.flowsery.com/mcp"

Then sign in with codex mcp login flowsery.

Gemini CLI

gemini extensions install https://github.com/Flowsery/agent

Hermes Agent

hermes plugins install flowsery
hermes plugins enable flowsery

Headless agents and clients without OAuth

Create an API token on the API tokens page and send it as a header. Keys start with flow_ws_.

{
  "mcpServers": {
    "flowsery": {
      "type": "http",
      "url": "https://mcp.flowsery.com/mcp",
      "headers": { "Authorization": "Bearer flow_ws_your_key" }
    }
  }
}

Example prompts

• "How's my traffic this week compared to last week?"
• "Which channels and campaigns brought paying customers this month?"
• "Visitors dropped on Tuesday. Check pages, referrers and devices and tell me why."
• "How many people are on the site right now, and where are they?"
• "How much traffic did ChatGPT, Perplexity and other AI assistants send us this month?"
• "What did Flowsery find in session recordings this week? Mark the checkout bug as in progress."
• "Show me the journey of the visitor who just started a free trial."
• "Which landing pages convert best for mobile visitors from Germany?"

Tools

28 tools. Every report takes a date range, a timezone and filters on country, region, city, device, browser, OS, page, hostname, referrer, channel and UTM parameters.

Workspaces and websites

Tool	What it does	Changes data
list_workspaces	List the workspaces the sign-in can act in; pass an id as workspaceId to any other tool	No
list_websites	List the websites the token can read, with domain, timezone, currency and KPI goal. Call it first	No
get_metadata	One website's settings: domain, timezone, currency, KPI goal, logo and colors	No

Traffic

Tool	What it does	Changes data
get_overview	Headline totals for a date range: visitors, sessions, bounce rate, session duration, revenue, revenue per visitor and conversion rate	No
get_timeseries	The same metrics by hour, day, week or month, with new, renewal and refunded revenue	No
get_realtime	Visitors active in the last 5 minutes	No
get_realtime_map	Active visitors with city-level location for a live map; coordinates rounded to 0.1 degrees, no IP addresses	No

Breakdowns

Tool	What it does	Changes data
get_pages	Page paths ranked by visitors	No
get_referrers	Referring sites ranked by visitors, with known sources such as Google, ChatGPT and Reddit named	No
get_channels	Traffic grouped into GA4-style channels: Organic Search, Paid Social, Email, Direct and more	No
get_campaigns	UTM campaigns ranked by visitors	No
get_countries	Visitors by country	No
get_regions	Visitors by region or state	No
get_cities	Visitors by city	No
get_devices	Desktop, mobile and tablet split	No
get_browsers	Visitors by browser	No
get_operating_systems	Visitors by operating system	No
get_hostnames	Visitors by hostname, for sites that track several domains or subdomains	No
get_breakdown	Group visitors by any of 25 dimensions, including entry page, exit link, browser and OS versions and every UTM parameter	No

Goals and visitors

Tool	What it does	Changes data
get_goals	Every goal, including the automatic payment and free trial goals, with completions in the window	No
track_goal	Record one completion of a custom goal; the goal is created on first use	Yes
delete_goals	Permanently delete goal completions that match the filters	Yes
track_payment	Record a payment for revenue attribution, or mark one refunded; never charges anyone	Yes
delete_payments	Permanently delete payment records that match the filters	Yes
get_visitor	One visitor's location, device, source, visits, goals, revenue and timeline	No

Session issues

Tool	What it does	Changes data
list_issues	Bugs, broken flows and UX problems the AI found in session recordings, ranked by severity	No
get_issue	One issue with its occurrences, sessions, steps to replicate, comments and linked Linear or Jira ticket	No
update_issue_status	Set an issue to open, in progress, resolved or suspended	Yes

Roles, permissions and safety

• Every call runs with the signed-in member's role in the workspace. Reads need workspace.read; tracking or deleting goals and payments and changing an issue's status need flowsery.write, which Admin and Editor hold.
• Read-only tools carry readOnlyHint. Only track_goal, delete_goals, track_payment, delete_payments and update_issue_status change anything, so clients that confirm writes ask before those five. Payment tools change analytics records only; they never charge, refund or touch a payment provider.
• Visitor tools return the analytics of your own websites. get_realtime_map rounds coordinates to 0.1 degrees, and no tool returns IP addresses or billing details.
• A website key (flow_) reaches one website; a workspace token (flow_ws_) or an OAuth sign-in reaches every website in the workspace.

Workspaces

An OAuth sign-in can reach several workspaces, each with its own data and your own role there. list_workspaces returns them, and every other tool takes an optional workspaceId, which the server sends as the X-Workspace-Id header. Without it a tool works in the workspace list_workspaces marks current. A workspaceId the sign-in cannot reach answers 403 with code: workspace_access_denied. An API key belongs to one workspace.

The skill

npx skills add Flowsery/agent

Or copy skills/flowsery/ into your project's skills folder. Cursor can also read it as a remote rule from https://raw.githubusercontent.com/Flowsery/agent/main/skills/flowsery/SKILL.md.

Setup

1. Create an account at flowsery.com, add your website and install the tracking snippet.
2. Create a workspace API token on the API tokens page.
3. Run:

   ./scripts/flowsery.js setup --key flow_ws_xxxxx
   

What the agent can do

• Headline totals for any date range: visitors, sessions, bounce rate, revenue and conversion rate
• Trends by hour, day, week or month
• Visitors active right now, with a live map
• Breakdowns by any of 25 dimensions: pages, landing pages, referrers, channels, UTM campaigns, countries, regions, cities, devices, browsers and more
• Filters on country, device, browser, UTM parameters, page and channel
• Goals and their completions, and recording custom goal events
• Recording payments for revenue attribution (Stripe, LemonSqueezy and Polar are tracked automatically)
• One visitor's full journey with their activity timeline
• The issues Flowsery's AI found in session recordings, and changing their status

You: How's my traffic this week?
Agent: Your site had 2,847 visitors and 3,912 sessions this week.
       Bounce rate is 62%. Revenue: $1,240 from 18 conversions.
       Top sources: Google (41%), Direct (28%), Twitter (12%).

Run the server yourself

The hosted server is the easy path. To run it yourself over stdio with Bun:

{
  "mcpServers": {
    "flowsery": {
      "command": "bun",
      "args": ["run", "/path/to/agent/mcp-server/src/index.ts"],
      "env": { "FLOWSERY_API_KEY": "flow_ws_your_key" }
    }
  }
}

Or in Docker:

docker build -t flowsery-mcp .
docker run -i -e FLOWSERY_API_KEY=flow_ws_your_key flowsery-mcp

bun run src/index.ts --http in mcp-server/ serves the same tools over Streamable HTTP on port 3100, the way https://mcp.flowsery.com/mcp runs. Environment variables and local development are in mcp-server/README.md.

FAQ

Do I need an API key? No. Point your client at https://mcp.flowsery.com/mcp and sign in with the email you use on Flowsery. Keys are for headless agents and clients without OAuth.

Does the agent see personal data? It sees what your Flowsery dashboard shows for your own websites: approximate location, device, source and activity. No IP addresses, and live map coordinates are rounded.

Can it change my analytics? Five tools write: track_goal, delete_goals, track_payment, delete_payments and update_issue_status. Everything else is read-only.

Does it work with several websites? Yes. Call list_websites and pass a websiteId or domain to the report tools.

What's the difference between this repo and Flowsery/claude-plugin? The plugin is the Claude Code and Hermes package: OAuth only, with a skill that drives these MCP tools. This repo has the server's source, the API-key skill for npx skills add and the Gemini extension.

Links

• Flowsery
• AI agents
• API tokens
• Claude Code and Hermes plugin
• Privacy policy: flowsery.com/privacy

License

MIT