frappe-cloud-mcp
An MCP server for the Frappe Cloud control plane that lets agents manage sites, bench groups, servers, backups, deploys, logs, analytics, and billing. It provides read, write, and optional destructive tools for operating Frappe Cloud hosting infrastructure.
README
frappe-cloud-mcp
An MCP server for the Frappe Cloud control plane — the press API behind
frappecloud.com. It gives an agent an operator's view of your hosting
account: sites, bench groups, servers, backups, deploys, logs, analytics and billing.
This is not another ERPNext data connector. Existing Frappe MCP servers talk to the DocTypes inside a site; this one talks to the layer that runs the site. Ask it why a deploy failed, what is hammering your database, or what next month's invoice looks like.
Install
Requires Python 3.11+.
git clone <this repo> && cd frappe-cloud-mcp
uv sync
Credentials
In the Frappe Cloud dashboard, go to Settings > Developer and press Create New API Key.
Despite the button's name it gives you two values — an API key and an API secret — and the secret is shown only once, so copy both before closing the dialog.
export FRAPPE_CLOUD_API_KEY=...
export FRAPPE_CLOUD_API_SECRET=...
Both halves are required. Frappe reads the header as token <api-key>:<api-secret> and splits it
on the colon; given only one value it discards the credential silently and serves the request as
an anonymous Guest, so a missing secret surfaces as Not Permitted rather than as a login error.
This server refuses to start on a half credential instead of letting you hit that.
If you have the pair as a single key:secret string, FRAPPE_CLOUD_TOKEN takes it whole.
Register with Claude Code
From this repo (project scope)
A .mcp.json ships in the repo root, so opening this project in Claude Code offers
the server automatically. It reads credentials from your environment rather than storing them, so
the file is safe to commit:
{
"mcpServers": {
"frappe-cloud": {
"type": "stdio",
"command": "uv",
"args": ["run", "--directory", "${CLAUDE_PROJECT_DIR:-.}", "frappe-cloud-mcp"],
"env": {
"FRAPPE_CLOUD_API_KEY": "${FRAPPE_CLOUD_API_KEY}",
"FRAPPE_CLOUD_API_SECRET": "${FRAPPE_CLOUD_API_SECRET}",
"FRAPPE_CLOUD_URL": "${FRAPPE_CLOUD_URL:-https://frappecloud.com}",
"FRAPPE_CLOUD_TEAM": "${FRAPPE_CLOUD_TEAM:-}",
"FRAPPE_CLOUD_READ_ONLY": "${FRAPPE_CLOUD_READ_ONLY:-0}",
"FRAPPE_CLOUD_ALLOW_DESTRUCTIVE": "${FRAPPE_CLOUD_ALLOW_DESTRUCTIVE:-0}"
}
}
}
}
Claude Code asks for approval the first time it sees a project-scoped server. Everything except
the key and secret has a :-default, so exporting those two is enough to start.
From anywhere (user scope)
To use it across all your projects, paste this into ~/.claude.json under mcpServers — with a
real absolute path, since ${CLAUDE_PROJECT_DIR} means nothing outside this repo:
{
"mcpServers": {
"frappe-cloud": {
"type": "stdio",
"command": "uv",
"args": ["run", "--directory", "/home/you/projects/frappe-cloud-mcp", "frappe-cloud-mcp"],
"env": {
"FRAPPE_CLOUD_API_KEY": "fc_key_...",
"FRAPPE_CLOUD_API_SECRET": "fc_secret_..."
}
}
}
}
The same block works in Claude Desktop's claude_desktop_config.json and in any other MCP client
that takes an mcpServers object. If uv is not on your PATH, use its absolute path
(which uv) as command.
Or add it from the CLI, without editing files:
claude mcp add frappe-cloud \
-e FRAPPE_CLOUD_API_KEY=... \
-e FRAPPE_CLOUD_API_SECRET=... \
-- uv run --directory /path/to/frappe-cloud-mcp frappe-cloud-mcp
Verify with /mcp in Claude Code: the server should show connected. Bad credentials show
failed with the HTTP status from Frappe Cloud.
Read-only or destructive
Both flags are ordinary env vars, so set them in the env block of whichever config you used:
"env": {
"FRAPPE_CLOUD_API_KEY": "...",
"FRAPPE_CLOUD_API_SECRET": "...",
"FRAPPE_CLOUD_READ_ONLY": "1"
}
A practical pattern is two entries — frappe-cloud read-only for everyday questions and
frappe-cloud-admin with FRAPPE_CLOUD_ALLOW_DESTRUCTIVE=1 for sessions where you intend to
change infrastructure.
Configuration
| Variable | Default | Purpose |
|---|---|---|
FRAPPE_CLOUD_API_KEY |
— | required |
FRAPPE_CLOUD_API_SECRET |
— | required |
FRAPPE_CLOUD_TOKEN |
— | alternative to the pair above, as key:secret |
FRAPPE_CLOUD_TEAM |
your default team | sent as X-Press-Team; needed to target a child team |
FRAPPE_CLOUD_URL |
https://frappecloud.com |
point at a self-hosted press instance |
FRAPPE_CLOUD_READ_ONLY |
0 |
expose read tools only |
FRAPPE_CLOUD_ALLOW_DESTRUCTIVE |
0 |
unlock the destructive tier |
FRAPPE_CLOUD_TIMEOUT |
60 |
request timeout in seconds |
Safety model
Tools fall into three tiers, and a tool that is not enabled is never registered — a read-only server truthfully advertises no mutating tools rather than offering them and refusing.
| Tier | Default | Examples |
|---|---|---|
| read | on | list_sites, site_log, bench_candidate, mariadb_processlist, upcoming_invoice |
| write | on (off under READ_ONLY) |
create_site_backup, clear_site_cache, install_app, add_site_domain |
| destructive | off unless ALLOW_DESTRUCTIVE |
archive_site, deploy_bench, change_site_plan, reboot_server, press_call |
FRAPPE_CLOUD_READ_ONLY=1 wins over FRAPPE_CLOUD_ALLOW_DESTRUCTIVE=1.
Run it read-only or default for day-to-day work; turn on the destructive tier only for a session where you actually intend to change infrastructure.
Tools
86 tools by default; 115 with the destructive tier on.
Sites — list/get, installed and available apps, config, domains, backups and download links, logs, agent jobs, activity, update checks, plans. Writes: backup, clear cache, install app, add domain, update config, activate, auto-update toggles. Gated: create, update, migrate, deactivate, archive, reinstall, restore, uninstall app, remove domain, change plan, version upgrade, move server, admin login link.
Bench groups — list/get, apps, config, dependencies, deployed versions, deploy information and status, recent deploys, deploy candidates (with build output), jobs, logs, processes, branches, regions. Writes: fetch latest app update, restart. Gated: deploy, update, rebuild, add/remove app, change branch, update config, archive.
Servers — list/get, overview, usage, bench groups, plans, jobs, Ansible plays, reclaimable disk. Gated: reboot, change plan, archive.
Observability — site analytics, daily usage, request count and duration by path, background job duration, request logs, server CPU/memory, MariaDB process list and slow queries, deadlock report, structured log browsing, notifications.
Billing (read-only throughout) — summary, upcoming invoice, current amount, unpaid and past invoices, balances, forecast, invoice line items, subscriptions.
Generic — press_methods searches a bundled catalogue of 278 whitelisted press endpoints;
press_list, press_get and press_search_link read arbitrary press documents. Gated:
press_call, press_run_doc_method and press_set_value reach anything without a named tool.
Refreshing the method catalogue
press_methods is backed by src/frappe_cloud_mcp/catalog.json, generated from the frappe/press
source. Regenerate it when press adds endpoints:
uv run python scripts/refresh_catalog.py --ref master
Development
uv run pytest # 44 tests, no network
uv run ruff check src tests scripts
uv run mcp dev src/frappe_cloud_mcp/__main__.py
Notes on the press API
Built against the frappe/press source rather than the published docs, which are stale and
currently returning 500s. Behaviour worth knowing:
- Responses wrap the return value in
{"message": ...}; errors arrive as 4xx/5xx with_server_messages, a JSON string holding a list of JSON strings. - Reads go out as GET with non-scalar arguments JSON-encoded per parameter; writes go as POST with a JSON body so argument types survive.
press.api.site.getanswers a custom domain with a redirect to the real site name; the client follows it.- The generic
press.api.client.*layer is limited server-side to an allowlist of ~55 doctypes and to documents owned by the active team.
Out of scope
OAuth/browser login, marketplace app publishing, partner and SaaS onboarding flows, data-plane access to the sites themselves, and remote HTTP transport (stdio only).
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.