Part A (the warm-up that proves the plumbing) gives no concrete way to launch a reference server — it says to pick one "your tool's docs point you at" and that they're "launched the same stdio way … a command and args," but never shows the command. Real reference servers (filesystem/fetch) are typically run via npx/uvx, requiring Node or uv, neither taught nor listed in "You'll need" (which lists only Python + the MCP SDK). So the learner can stall on the easy warm-up.
Evidence
modules/20-mcp-servers-giving-the-ai-hands/README.md (~lines 249-259): "pick a simple, read-only one your tool's docs point you at (a 'filesystem' or 'fetch' server …)" and "Most reference servers are launched the same stdio way … a command and args." No runtime prereq. (Verify-before-publish ~lines 421-423 deliberately avoids naming a specific server.)
Why it matters
A Python-only learner can stall on a self-described warm-up before reaching Part B (which lands the concept on the already-installed SDK) — against the honesty/labs-must-run promises.
Proposed change
Make Part A explicitly optional, and add one tool-agnostic line to "You'll need": "the reference server your tool points you at may run via npx (Node) or uvx (uv) — install whichever its documented command needs."
This closes the honesty gap without hard-pinning a specific server command (consistent with the module's deliberate stance) and without forcing a Node install for a 10-minute warm-up. Adjust the Check-for-understanding/learning-objective wording slightly since Part C still connects a server.
Acceptance criteria
Part A names the possible runtime prereqs (Node/uv) tool-agnostically, or is clearly optional.
The load-bearing concept path (Part B/C) remains runnable with the already-listed Python SDK.
Filed from an adversarial multi-agent course review (217 raw findings → 54 adversarially-verified survivors). Scoped for manual review; intentionally not auto-assigned to an agent.
## Problem
Part A (the warm-up that proves the plumbing) gives no concrete way to launch a reference server — it says to pick one "your tool's docs point you at" and that they're "launched the same stdio way … a command and args," but never shows the command. Real reference servers (filesystem/fetch) are typically run via `npx`/`uvx`, requiring Node or uv, neither taught nor listed in "You'll need" (which lists only Python + the MCP SDK). So the learner can stall on the easy warm-up.
## Evidence
`modules/20-mcp-servers-giving-the-ai-hands/README.md` (~lines 249-259): "pick a simple, read-only one your tool's docs point you at (a 'filesystem' or 'fetch' server …)" and "Most reference servers are launched the same stdio way … a command and args." No runtime prereq. (Verify-before-publish ~lines 421-423 deliberately avoids naming a specific server.)
## Why it matters
A Python-only learner can stall on a self-described warm-up before reaching Part B (which lands the concept on the already-installed SDK) — against the honesty/labs-must-run promises.
## Proposed change
1. Make Part A explicitly optional, and add one tool-agnostic line to "You'll need": "the reference server your tool points you at may run via `npx` (Node) or `uvx` (uv) — install whichever its documented command needs."
This closes the honesty gap without hard-pinning a specific server command (consistent with the module's deliberate stance) and without forcing a Node install for a 10-minute warm-up. Adjust the Check-for-understanding/learning-objective wording slightly since Part C still connects a server.
## Acceptance criteria
- [ ] Part A names the possible runtime prereqs (Node/uv) tool-agnostically, or is clearly optional.
- [ ] The load-bearing concept path (Part B/C) remains runnable with the already-listed Python SDK.
## Affected files
- `modules/20-mcp-servers-giving-the-ai-hands/README.md`
## References
Source finding F33 (realVotes 3/3).
---
*Filed from an adversarial multi-agent course review (217 raw findings → 54 adversarially-verified survivors). Scoped for manual review; intentionally not auto-assigned to an agent.*
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Problem
Part A (the warm-up that proves the plumbing) gives no concrete way to launch a reference server — it says to pick one "your tool's docs point you at" and that they're "launched the same stdio way … a command and args," but never shows the command. Real reference servers (filesystem/fetch) are typically run via
npx/uvx, requiring Node or uv, neither taught nor listed in "You'll need" (which lists only Python + the MCP SDK). So the learner can stall on the easy warm-up.Evidence
modules/20-mcp-servers-giving-the-ai-hands/README.md(~lines 249-259): "pick a simple, read-only one your tool's docs point you at (a 'filesystem' or 'fetch' server …)" and "Most reference servers are launched the same stdio way … a command and args." No runtime prereq. (Verify-before-publish ~lines 421-423 deliberately avoids naming a specific server.)Why it matters
A Python-only learner can stall on a self-described warm-up before reaching Part B (which lands the concept on the already-installed SDK) — against the honesty/labs-must-run promises.
Proposed change
npx(Node) oruvx(uv) — install whichever its documented command needs."This closes the honesty gap without hard-pinning a specific server command (consistent with the module's deliberate stance) and without forcing a Node install for a 10-minute warm-up. Adjust the Check-for-understanding/learning-objective wording slightly since Part C still connects a server.
Acceptance criteria
Affected files
modules/20-mcp-servers-giving-the-ai-hands/README.mdReferences
Source finding F33 (realVotes 3/3).
Filed from an adversarial multi-agent course review (217 raw findings → 54 adversarially-verified survivors). Scoped for manual review; intentionally not auto-assigned to an agent.