pipefy/ai-toolkit

Overview • Installation • What it connects to • MCP server • CLI • Agent skills • Documentation • Contributing • Legal

---

Overview

Connect Claude Code, Cursor, and other AI agents to Pipefy.

The toolkit has an MCP server, a CLI, and a GraphQL SDK. Those three call Pipefy's API with the signed-in account's permissions. Agent skills cover common Pipefy workflows.

Component	Package / path	Purpose
MCP server	pipefy-mcp-server	Pipefy tools for MCP clients (Claude Code, Cursor, Claude Desktop, and others). Hosted at https://mcp.pipefy.com/mcp, or run locally for the full catalog.
CLI	pipefy-cli	Most MCP capabilities as terminal commands, for scripting and CI. Coverage and MCP-only tools: docs/parity.md.
SDK	pipefy	GraphQL client, services, and models shared by the MCP server and the CLI.
Skills	plugins/pipefy/skills/	Markdown playbooks (Anthropic Skills format) for common Pipefy workflows.

• Feedback: GitHub Issues
• Email: [email protected]

---

Installation

Pick the row for your client and run its command. Flags, caveats, and the remaining steps are in docs/install.md.

You are in	Path	Start with
Claude Code, fastest start	Hosted MCP	claude mcp add --transport http --scope user --client-id pipefy-mcp pipefy https://mcp.pipefy.com/mcp
Claude Code, with skills, slash commands, and the CLI	Claude Code plugin	/plugin install pipefy --marketplace pipefy/ai-toolkit for Claude Code 2.1.275 or later. The install guide has the two-step form for older versions. /pipefy:install adds the CLI. /pipefy:pipefy-login signs it in.
Cursor	Cursor Marketplace plugin	Install Pipefy from the Cursor Marketplace
Cursor, Claude Desktop, or Codex on macOS or Linux, when you need the local-file tools or the CLI	Quick-install script	Run the install script with --client cursor, claude-desktop, or codex
A terminal, a script, or CI	CLI only	uv tool install pipefy-cli, then pipefy auth login
Any Markdown-aware agent, playbooks only	Skills only	npx skills add pipefy/ai-toolkit

• One MCP server. Register exactly one Pipefy MCP server at a time. Use the hosted URL, a local server, or a plugin. The scan command reports what a machine already has. It changes nothing. To remove or switch paths, see docs/uninstall.md.
• Sign-in. The hosted server uses your client's OAuth prompt. Local paths use pipefy auth login (browser OAuth), a service account for unattended use, or a PIPEFY_TOKEN. See docs/config.md and docs/cli/auth.md.

---

What it connects to

• Pipefy hosts and storage URLs. The toolkit calls Pipefy hosts and the storage URLs Pipefy issues.
• Client headers. Requests carry a client name and version. The toolkit sends no other telemetry.
• What an install can add. An install can add uv, the CLI, a local MCP server, one client config entry, skills, and a keychain session.
• Third-party skills telemetry. The skills tool sends its own telemetry unless the environment contains DISABLE_TELEMETRY or DO_NOT_TRACK.

Hosts, headers, and files: docs/trust.md.

---

MCP server

• Local server. It registers the full catalog. Canonical names are PIPEFY_TOOL_NAMES in registry.py.
• Hosted server. https://mcp.pipefy.com/mcp serves the remote-safe surface. It omits tools that read a file on your machine. Those tools are knowledge-base document upload, and custom LLM-provider create and update.
• Attachments on the hosted server. Attachment tools take a file_url instead of a file_path.
• Tool reference: docs/mcp/
• Shared conventions: docs/mcp/tools/cross-cutting.md
• Narrow the local server: Choosing a tool surface

---

Command-line interface

• The pipefy CLI mirrors the MCP capabilities that docs/parity.md marks as shipped.
• Most commands print Rich tables by default and JSON with --json.
• On destructive commands, --yes skips the prompt.
• Without --yes, interactive confirmation still runs.
• pipefy graphql exec refuses a mutation unless the command includes --yes.

pipefy pipe list --json
pipefy card get 123456789
pipefy introspect query pipe

• Guides: docs/cli/ and packages/cli/README.md

---

Agent skills

• plugins/pipefy/skills/ holds workflow playbooks for any agent that reads Markdown.
• Each playbook has prerequisites, a tool table, steps, and success criteria.
• Catalog: plugins/pipefy/README.md
• Authoring guide: plugins/AGENTS.md

---

Documentation

Document	Description
docs/install.md	Every install path in full, switching between them, tool-surface selection, and testing the Claude Code plugin from a checkout.
docs/trust.md	Hosts the toolkit calls, request headers, and files an install or login writes.
docs/README.md	Index of docs by surface (MCP, CLI, SDK) and shared guides.
docs/config.md	PIPEFY_* environment variables, config.toml schema and path, precedence chain.
docs/parity.md	MCP tool ↔ CLI command matrix.
docs/uninstall.md	uninstall.sh --scan and teardown, and switching between the hosted, local, and plugin channels.
docs/MIGRATION.md	Notes for existing MCP users. Configuration remains compatible.
AGENTS.md	Repository layout, development workflow, and guidelines for contributors and agents.
RELEASE.md	Versioning and release process.

---

Contributing

You can contribute through an issue or a pull request. Each commit needs a DCO sign-off (git commit -s).

• Skills. Playbook rules, frontmatter, and review steps are in CONTRIBUTING.md.
• MCP, CLI, and SDK. Repository layout, tests, and new capabilities are in AGENTS.md.
• Field types. If the tools do not map a field type, open an issue with that type and the behavior you expect.

---

Legal

This toolkit is under the Apache License 2.0, and use of Pipefy is governed by Pipefy's terms. Read TERMS.md for the full notice.

• License. The MCP server, the CLI, the SDK, and the skill and blueprint playbooks are under that license. Pipefy keeps its trademarks. See NOTICE.
• Platform terms. Your use of the Pipefy platform through this toolkit is governed by the Pipefy Solutions Terms and Conditions (versão em português). AI features also follow the Pipefy AI Additional Terms. Those terms are not published yet, and they will be incorporated by reference into the platform terms, including their acceptable use provisions. You need a Pipefy account. The Apache 2.0 license grants no rights to the Pipefy service.
• Templates. Skills and blueprints are configurable templates for general informational purposes. They are not legal, HR, financial, tax, or other professional advice. Qualified professionals must review, configure, and validate them before production use. Each published blueprint ships with a COMPLIANCE.md that states the intended purpose, out-of-scope uses, AI risk classification, and default human-oversight settings. See docs/compliance/COMPLIANCE.template.md.
• AI output. Outputs from AI features may contain errors or omissions. Review each output on your own before you rely on it.
• Beta. Pre-1.0 releases are provided "AS IS", without warranties of any kind. They may change or be discontinued at any time.

Report security issues through SECURITY.md. Community standards are in CODE_OF_CONDUCT.md.