Guide · Agent skill
How to use the Prepare MCP Integration agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
Prepare MCP Integration is an agent skill for adding a new third-party service, through MCP (the Model Context Protocol) or a vendor API, to the Paperclip platform. It is written for Paperclip maintainers and engineers who want research and human approval recorded before any connector code is written. After this guide you will know what the skill makes the agent do, how to install it, how to ask for it in plain language, and what it needs before it runs.
Prepare MCP Integrationpaperclipai/paperclip · MIT Download the skill Quick start
- 1Download and unzip the skill. Get the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills.
- 2Confirm prerequisites. Make sure git, gh and curl are available and that you can access the paperclipai/paperclip and paperclip-content repositories.
- 3Gather the vendor source. Collect the vendor documentation URL and any brief, and identify the intended MCP endpoint or API.
- 4Ask for research only. Send a request naming the service and stating that no connector should be built until you approve the research.
- 5Review the research PR. Check the cited claims, marked unknowns and proposed connection split in the paperclip-content PR.
- 6Confirm the exact revision. Approve the specific research head SHA and the connection set you want, via the confirmation request.
- 7Review each connector PR. After approval, review the separate Paperclip App PR for each approved connection.
What the skill makes the agent do
The skill splits the work into two phases. In Phase A the agent does cited research and opens a documentation-only pull request in the paperclip-content repository for human review. It does not create any Paperclip App branch, task or code change at that stage.
In Phase B, which starts only after a person accepts the exact research revision, the agent builds connector code. It opens one separate pull request per approved connection, so distinct connections never share a connector PR.
- Phase A: cited research, a research-only PR, and a confirmation request against a specific head SHA.
- Phase B: connector implementation in the Paperclip App, one PR per approved connection.
- The skill points to other repository files (the integration README, the integration harness, and CONNECTOR-PLAYBOOK.md) as the authoritative rules rather than containing those rules itself.
Install the skill (ZIP download or Claude apps upload)
Download the skill ZIP from this page and unzip it. For a personal install, place the unzipped folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder.
If you use the Claude apps instead, upload the skill under Settings → Capabilities → Skills.
The skill folder includes SKILL.md, which holds the instructions, and an example file, examples/notion-mcp-research-gate.md, that shows a full run.
- Personal install: ~/.claude/skills/
- Project install: .claude/skills/ inside the project
- Claude apps: Settings → Capabilities → Skills
Check the requirements before you trigger it
The skill declares three tools: git, gh (the GitHub CLI) and curl. It also needs access to the paperclipai/paperclip and paperclip-content repositories, because it reads files from them and depends on their playbooks. Without that access the skill cannot run correctly, since its rules live outside this folder.
You also need vendor documentation and, when credentials are involved, approved secret storage for vendor credentials. The skill is not meant for production vendor accounts without review.
- git, gh and curl available to the agent
- Access to paperclipai/paperclip and paperclip-content
- Vendor documentation for the service you are adding
- Approved secret storage if vendor credentials are needed
Trigger the skill with a natural request
Describe the integration and say that you want research first and no build until you approve it. The skill is designed for new integration research and delivery. Its description says it is not for ad hoc connector coding that bypasses the playbooks, so ask for the research gate explicitly.
The example file shows the pattern: give the vendor documentation URL, name the service, and state that the connector must not be built until you approve the research.
Research Notion's hosted MCP server and prepare it for Paperclip. Start from the vendor documentation URL. Do not build the connector until I approve the research.Review and approve the research gate
When Phase A finishes, the agent records the research PR URL, its exact head SHA, the content and App source commits used, and the proposed connection set in an issue document. It then asks you to confirm against that exact revision.
Approval covers only that commit and only the connections you accepted. If the PR gets new commits or you want an extra connection, the approval does not carry over and the research needs to be confirmed again.
- Check that research claims include URLs and access dates.
- Check that unknowns are marked, not guessed.
- Approve only the connection set you actually want built.
Understand the safety notes
The skill does not run scripts, but it does use the network. The research relies on vendor documentation and safe live probes. The skill says not to mutate a vendor account, register a client, grant consent or invoke a write tool just to research it. Credential- or browser-dependent validation goes to an explicit QA task.
Credentials must stay in approved secret storage. The skill forbids putting secrets in briefs, catalog files, issue text, plans, fixtures, screenshots, logs, branch names, commits or PRs.
- Runs scripts: no. Uses network: yes.
- Unauthenticated metadata probes are allowed only if they do not change vendor state.
- Vendor permission breadth is separate from Paperclip execution governance, so review actions, approvals and audit policies on their own terms.
What people use it for
- Researching a hosted vendor MCP server, such as Notion, before anyone writes connector code.
- Splitting one vendor brief into several independently reviewable connections, each with its own PR.
- Recording research evidence, URLs and access dates in a PR that a maintainer can review.
- Adding a prerequisite or playbook correction in its own PR before the connector work begins.
Frequently asked questions
How do I install the prepare-mcp-integration skill?
Download the skill ZIP from this page and unzip it into ~/.claude/skills/ for a personal install, or into a project's .claude/skills/ folder. In the Claude apps you can instead upload it under Settings → Capabilities → Skills.
What do I need to run it?
The skill declares git, gh (GitHub CLI) and curl. You also need access to the paperclipai/paperclip and paperclip-content repositories, vendor documentation, and approved secret storage if vendor credentials are involved.
Is the prepare-mcp-integration skill free?
The source material provided for this page does not state a price or licence terms for the skill itself. The catalog entry lists the licence as MIT, so check the repository's licence file for the exact terms.
Does the skill write code right away?
No. It first does research and opens a documentation-only PR. Connector code is written only after a person approves that exact research revision and connection set.
Can I use it for ad hoc connector coding?
The skill is not meant for that. Its description says it is not for ad hoc connector coding that bypasses the playbooks, and it warns against skipping the playbooks.
Does the skill run scripts or touch the network?
It does not run scripts, but it does use the network for vendor documentation and safe live probes. It avoids changing vendor accounts, registering clients or granting consent during research.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you maintain Paperclip or add vendor integrations and want research, citations and human approval recorded before any connector code exists. Skip it if you need quick ad hoc connector coding, or if you do not have access to the paperclip and paperclip-content repositories, since the skill relies on files outside its own folder.
Download the skill