mcp-server-offering Skill (samber): Design Your Product's MCP Server
Makes Claude design the MCP server your product offers to AI agents: which tools to expose, how to protect writes, and how to roll it out.
- Installs
- 9.6K
- on skills.sh
- Stars
- 9
- on GitHub
- Files
- 5
- besides SKILL.md
- Updated
- Nov 2026
- last repo commit
mcp-server-offering makes Claude design a SaaS product's MCP server as a product surface for AI agents. It sizes the investment, curates 5 to 15 workflow tools instead of mirroring API endpoints, applies write-tool safety patterns, picks hosting and sign-in, versions the tools and plans a staged rollout.
It starts when you mention MCP, an agent-facing tool surface, or exposing a product to AI assistants. Claude asks 6 opening questions, then works through 8 steps. SKILL.md calls it the design layer: code-generation MCP builder skills scaffold the server once you know what it should expose.
Good for
- Product and platform teams asked by customers for agent access
- Engineers whose MCP server mirrors 60 endpoints and confuses agents
- Leads deciding how much to invest after a competitor shipped an MCP server
Install mcp-server-offering
Claude Code
- 1
Copy this prompt.
Install the mcp-server-offering skill in Claude Code. Run this command: /plugin marketplace add samber/developer-platform-skills /plugin install developer-platform-skills@developer-platform-skills Then check that it loaded and tell me how to use it.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Pick one route, not both, or you get every skill twice. The README asks you to install the whole set, not just mcp-server-offering, because the skills reference each other and a lone skill leaves those hand-offs with nowhere to go. The plugin adds all 61 skills in the repo (counted from the repo tree). Its manifests list no hooks, agents or MCP servers.
From the project README. Source.
What's in the skill
When it loads
Design a SaaS product's MCP server as a product surface AI agents operate - sizing the build investment, curating a 5-15 workflow-tool agent surface instead of mirroring API endpoints, MCP Apps interactive UI, named write-tool safety patterns (scoped credentials, read-only lockdown, human-in-the-loop approval, risk-tiered server-side gates, dry-run preflight, idempotency and spend caps), remote hosting with OAuth 2.1, versioning the tool surface, MCP registry discoverability, and measuring agent adoption. Use whenever the user mentions MCP, Model Context Protocol, an agent-facing tool surface, exposing a product to AI agents or coding assistants, or MCP write-tool safety - even if they never say "MCP server". Design layer only - code-generation MCP builder skills scaffold what this specifies.
Sections in SKILL.md
- Clarifying questions
- Audience split: agent-consumer type
- Workflow
- 1. Size the build investment
- 2. Curate the tool surface
- MCP Apps: when a tool returns interactive UI
- 3. Apply write-tool safety patterns
- 4. Choose hosting and auth
- 5. Version the tool surface
- 6. Publish for discoverability
- 7. Measure agent adoption
- 8. Roll out in stages
- Failure modes
- Measurement
- Invocation examples
- References
Key facts
- Author
- Samuel Berthe
- License
- MIT
- Version
- 1.1.3
- Installs on skills.sh
- 9,563 (Nov 2, 2026)
- GitHub stars
- 9 (Nov 2, 2026)
- skills.sh audits
- Gen Agent Trust Hub pass; Socket pass; Snyk pass (Nov 2, 2026)
- Last repo commit
- Nov 1, 2026
What it includes
Asks 6 questions in one batch before designing anything.
Try: “We have a stable REST API and customers asking for agent access - design our MCP server: which tools do we expose and how do we secure the write ones?”
Opening questionsSets how much to invest in an MCP server and in what order.
Size the build investmentDesigns 5 to 15 workflow tools, not a copy of every API endpoint.
Curate the tool surfaceOrders six patterns for tools that change data, such as scoped credentials and approval.
Apply write-tool safety patternsRanks three hosting options and covers sign-in with OAuth 2.1.
Choose hosting and authVersions three layers separately, starting with the protocol version.
Version the tool surfaceLists the server in the official MCP registry and in third-party directories.
Publish for discoverabilityTracks call volume, errors, per-tool use, active users and client mix.
Measure agent adoptionStarts with a read-only launch of the curated tools, then widens on measured thresholds.
Roll out in stagesAdvises saving decisions to developer-platform-context.md when memory is a file.
Write ·
Remember decisions (optional)
From the official docs
Permissions
Low risk
This skill is written advice that bundles no scripts. The one thing it may write is a decisions file, so check what Claude saves.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
It may save decisions to a file
SKILL.md advises using developer-platform-context.md when memory lives in a file, and calls it an advisory reference. Check or delete that file if you do not want decisions kept. github.com
The plugin route adds all 61 skills
The README asks you to install every skill in the repo, and the repo has 61. The plugin manifests list no hooks, agents or MCP servers, and the repo tree holds none. github.com
Folder holds Markdown and one JSON file
This skill's folder has 6 files: SKILL.md, evals/evals.json and 4 reference files. It has no .py, .sh or .js file. github.com
Security advisories
No published security advisories as of Nov 2, 2026.
Technical details
| Turning on read-only | No read-only setting is documented. Decline the memory-file suggestion if you want no files written. |
|---|---|
| Scopes |
|
Questions
How do I install mcp-server-offering?
Use the install box above. The README asks you to install all 61 skills in the repo, not just this one, because the skills point to each other. In Claude Code that is the plugin route, and elsewhere the skills CLI or the client's own route installs the set.
Does mcp-server-offering work in Claude.ai, Codex or Cursor?
The README documents installs for Claude.ai, Claude Code, Codex, Cursor and Gemini CLI. It also lists the skills CLI, which it calls universal for Agent Skills-compatible tools. The Claude.ai route adds the whole repo as a plugin marketplace.
Does it need an API key or any setup?
SKILL.md sets no environment variable and the folder bundles no scripts. It does ask for details of your own API or product in its opening questions.
Is mcp-server-offering safe to use?
The folder holds Markdown and one JSON file, and no scripts. SKILL.md does suggest saving settled decisions to developer-platform-context.md when your memory lives in a file, so check what Claude writes. Our advice: treat the design it produces as a draft for your team to review before you publish it.
mcp-server-offering vs api-integration-surface-strategy?
api-integration-surface-strategy decides whether to offer MCP at all, and this skill starts once MCP is on the roadmap. SKILL.md also points to cli-interface-design, ide-extension-offering, agent-skill-offering and agent-integration-surfaces for the neighbouring surfaces.
Does it build the MCP server for me?
It designs what the server should expose. SKILL.md calls it the design layer and says MCP builder skills and generators scaffold the code afterwards. Whether to offer MCP at all is decided by api-integration-surface-strategy, and this skill starts once MCP is on the roadmap.
Information
updated Nov 2, 2026
- Repository
- samber/developer-platform-skills
- Folder
skills/mcp-server-offering
Sources
Docs, issue trackers and discussions this page is built from.
Show all 6Hide
- README · github.com
- mcp-server-offering SKILL.md · github.com
- mcp-server-offering folder · github.com
- api-integration-surface-strategy SKILL.md · github.com
- marketplace.json · github.com
- plugin.json · github.com