- Home
- Skills
- APIs & Backend
- MCP Server Architect for AI Coding Agents
More screenshots
Works with the AI Agents you already use
MCP Server Architect for AI Coding Agents
Most MCP servers fail the same five checks — no Resource Server metadata, unpinned tools, prose-narrating returns. Design and review yours like a production system.
$15
MCP Server Architect for AI Coding Agents
Example session with this skill installed
Review my MCP server implementation. It uses stdio transport and provides tools for interacting with a local Postgres database. I need a security audit and spec compliance check.
- Read your context and instructions
- Compiled the mcp server architect
I have analyzed the provided MCP server implementation for the Postgres local domain. The server uses stdio transport, which is appropriate for this local-only use case.
MCP Design Review — postgres-local-server
Verdict: FIX-BEFORE-SHIP
Spec compliance matrix
| Area | Requirement | Status | Note |
|---|---|---|---|
| Transport | stdio | PASS | Correct for local single-user tool. |
| Auth | Stateless | PASS | No session state identified in core dispatch. |
| Tools | MRTR Pattern | FAIL | query_db returns prose narration instead of raw JSON. |
| Tools | Hash-pinning | FAIL | No definition hashing detected; vulnerable to poisoning. |
| Listings | Cacheable | PASS | ETag support implemented for tools/list. |
Findings (by severity)
High / Tools / MRTR Violation: The query_db tool returns "Found X rows: [data]" which forces the model to parse prose.
Fix: Return raw JSON result set and count.
High / Security / Tool Poisoning: Tool definitions are loaded without integrity checks.
Fix: Implement hash-pinning at registration; block execution on mismatch.
- Med / Naming / snake_case: Tool
RunQueryuses PascalCase. Fix: Rename torun_query.
Tool surface audit
| tool | scopes | MRTR ok? | naming ok? | poisoning-safe? |
|---|---|---|---|---|
| get_tables | N/A | YES | YES | NO |
| query_db | N/A | NO | YES | NO |
Architecture notes
The server correctly identifies as a local stdio transport. Since this is stdio, OAuth is skipped, but the lack of tool integrity checks remains a critical risk for local execution.
Ship checklist
- Refactor
query_dbto return structured JSON only. - Implement SHA-256 hash validation for tool definitions.
- Pass MCP Inspector locally.
- Generate
claude_desktop_config.jsonsnippet.
Next steps
- Apply the MRTR fix to the
query_dbfunction to ensure clean model interpretation. - Add the hash-pinning logic to the server bootstrap sequence.
- Run the MCP Inspector and provide the logs for final verification.
Connects securely to your tools. The creator never sees your data.
About this skill
About this skill
MCP is the integration layer between agents and real tools — which makes a badly designed server an attack surface with a friendly JSON API. This skill encodes the full production architecture: stateless core, correct transport, OAuth 2.1 Resource Server with CIMD, per-tool scopes, tool-poisoning defense, and dual-era protocol support — plus a review method that catches the failures everyone actually ships.
What it does
- Enforces one-server-one-domain boundaries and correct transport selection
- Designs OAuth 2.1 correctly: protected-resource metadata, CIMD, PKCE, DPoP
- Audits tool surfaces: naming, MRTR structured returns, per-tool scopes, hash-pinning
- Plans dual-era protocol support with a deprecation path
- Outputs a compliance matrix + ship checklist (Inspector, registry, CVE watch)
Why this beats prompting it yourself
"Design an MCP server" produces a tutorial-shaped answer: a couple of tools and maybe auth. It will not mention tool-definition pinning, MRTR violations, removed session headers, or CIMD — because those live in the gap between "it connects" and "it survives production". This skill is that gap, codified.
Does not build client-side agent logic. Focuses exclusively on server-side architecture. Requires external verification for spec changes released after the skill's last update.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 10 days ago
Requires
Permissions
No special permissions declared or detected