docs: add mcp-cli usage to agent instructions

.mega-linter.yml

@@ -9,6 +9,7 @@ DISABLE_LINTERS:
   - PYTHON_PYLINT
   - PYTHON_PYRIGHT
   - SPELL_CSPELL
+  - REPOSITORY_SEMGREP
   - TERRAFORM_TERRASCAN
 DISABLE_ERRORS_LINTERS:
   - COPYPASTE_JSCPD

chezmoi/private_dot_agents/AGENTS.md

@@ -2,56 +2,54 @@
 
 ## Task Workflow
 
-- Read full context before coding:
-  README, AGENTS.md, related files.
+- Read full context before coding: README, AGENTS.md, related
+  files.
 - Work incrementally. Verify each step.
-- Confirm before destructive actions
-  (delete, force-push, overwrite).
+- Confirm before destructive actions (delete, force-push,
+  overwrite).
 - Don't modify files outside task scope.
-- Scope searches narrowly. Don't read entire
-  codebase — target specific files and symbols.
-- Ask one concise question when unclear.
-  Don't guess.
+- Scope searches narrowly. Don't read entire codebase — target
+  specific files and symbols.
+- Ask one concise question when unclear. Don't guess.
 
 ## Context Management
 
 - Clear context between unrelated tasks.
-- Use subagents for large investigations
-  to keep main context clean.
-- Don't let exploration fill context —
-  scope it, summarize findings, move on.
+- Use subagents for large investigations to keep main context
+  clean.
+- Don't let exploration fill context — scope it, summarize
+  findings, move on.
 
 ## Documentation
 
-- Read the project's `README.md` and agent
-  instruction file (`CLAUDE.md`, `GEMINI.md`,
-  or `AGENTS.md`) at the start of every task.
+- Read the project's `README.md` and agent instruction file
+  (`CLAUDE.md`, `GEMINI.md`, or `AGENTS.md`) at the start of
+  every task.
 - Keep both files updated when changes affect them.
-- `README.md` owns: project overview,
-  setup instructions, architecture, and usage docs.
-- Agent instruction file owns: AI agent workflow,
-  project-specific commands, and conventions.
-- Do not duplicate information between the two files.
-  Reference `README.md` for project context.
+- `README.md` owns: project overview, setup instructions,
+  architecture, and usage docs.
+- Agent instruction file owns: AI agent workflow, project-specific
+  commands, and conventions.
+- Do not duplicate information between the two files. Reference
+  `README.md` for project context.
 
 ## Code Quality
 
 - Follow existing patterns and conventions in codebase.
-- Write self-explanatory code. Comment only WHY,
-  never WHAT.
+- Write self-explanatory code. Comment only WHY, never WHAT.
 - No redundant, obvious, or outdated comments.
-- Use clear naming over comments.
-  Best comment is one you don't need.
+- Use clear naming over comments. Best comment is one you don't
+  need.
 - Keep changes minimal. No unrelated cleanup.
-- No premature abstractions. Three similar lines
-  beat a premature helper.
+- No premature abstractions. Three similar lines beat a premature
+  helper.
 
 ## Testing
 
 - Add tests for new logic and bug fixes.
 - Run existing tests before committing.
-- When bug found, capture root cause as a code
-  invariant or test case so same class of mistake never recurs.
+- When bug found, capture root cause as a code invariant or test
+  case so same class of mistake never recurs.
 - Don't skip failing tests — fix or flag them.
 
 ## Dependencies
@@ -60,52 +58,58 @@
 - Minimize additions. Justify when necessary.
 - Pin versions explicitly.
 
+## MCP Servers
+
+- Access MCP servers via the `mcp-cli` CLI when no native MCP
+  integration is available.
+- Run `mcp-cli --help` for commands, config search paths, and
+  examples.
+- Inspect a tool's schema using `mcp-cli info <server> <tool>`.
+- Execute tools using `mcp-cli call <server> <tool> '<json_args>'`.
+- Prefer native MCP tools over `mcp-cli` when both exist.
+
 ## Security
 
 - Never hardcode secrets, tokens, or credentials.
-- Validate inputs at system boundaries
-  (user input, external APIs).
-- Use parameterized queries. No string concatenation
-  for SQL/commands.
+- Validate inputs at system boundaries (user input, external
+  APIs).
+- Use parameterized queries. No string concatenation for
+  SQL/commands.
 - Don't commit `.env`, credentials, or key files.
 - Flag security concerns immediately when spotted.
 
 ## Git Workflow
 
-- Work in a feature branch,
-  not the default branch.
-- Branch naming: `<type>/<short-description>`
-  (e.g., `feat/add-auth`, `fix/login-redirect`).
+- Work in a feature branch, not the default branch.
+- Branch naming: `<type>/<short-description>` (e.g.,
+  `feat/add-auth`, `fix/login-redirect`).
 - Run project formatter before committing.
-- Make commits regularly as you progress —
-  don't accumulate large uncommitted changes.
+- Make commits regularly as you progress — don't accumulate large
+  uncommitted changes.
 - Follow [Conventional Commits](https://www.conventionalcommits.org/)
   for messages:
   - Format: `<type>(<scope>): <description>`
-  - Types: feat, fix, refactor, docs,
-    test, chore, ci, build, perf, style, revert
-  - Scope is optional but encouraged
-    when it adds clarity.
-  - Subject line ≤ 50 characters, imperative mood
-    ("add", "fix", "remove" — not "added", "adds").
+  - Types: feat, fix, refactor, docs, test, chore, ci, build,
+    perf, style, revert
+  - Scope is optional but encouraged when it adds clarity.
+  - Subject line ≤ 50 characters, imperative mood ("add", "fix",
+    "remove" — not "added", "adds").
   - No trailing period on subject line.
   - Body only when "why" isn't obvious from subject.
-  - Never write "This commit does X",
-    "I", "we" in body — diff says what.
-- Push the feature branch regularly
-  so work isn't lost.
-- When task complete, open a pull request
-  against the default branch.
+  - Never write "This commit does X", "I", "we" in body — diff
+    says what.
+- Push the feature branch regularly so work isn't lost.
+- When task complete, open a pull request against the default
+  branch.
 
 ## Code Review
 
-- Severity levels: critical (blocks merge),
-  high, medium, low (non-blocking).
+- Severity levels: critical (blocks merge), high, medium, low
+  (non-blocking).
 - Be specific: reference file paths and line numbers.
 - Suggest fixes, not just problems.
 - Check: correctness, security, tests, performance.
-- Approve only when all critical and high
-  items are resolved.
+- Approve only when all critical and high items are resolved.
 - Run tests locally before approving.
 
 ## Caveman Mode
@@ -114,37 +118,31 @@ Answer fast, use minimal words, no fluff.
 
 ### Core Directives
 
-- **Terse Output**: One sentence max per thought.
-  No elaboration unless asked.
-  Target 50-70% fewer tokens than normal mode.
-- **Structure**: Bullets, short code blocks, tables.
-  No prose paragraphs.
-  No greetings, summaries, meta-commentary.
-- **Word Budget**: Answer in fewest words that convey
-  meaning. Trim every sentence.
-- **Code Same**: Code output is standard
-  (readable, well-formatted).
-  Only chat responses are terse.
+- **Terse Output**: One sentence max per thought. No elaboration
+  unless asked. Target 50-70% fewer tokens than normal mode.
+- **Structure**: Bullets, short code blocks, tables. No prose
+  paragraphs. No greetings, summaries, meta-commentary.
+- **Word Budget**: Answer in fewest words that convey meaning.
+  Trim every sentence.
+- **Code Same**: Code output is standard (readable,
+  well-formatted). Only chat responses are terse.
 
 ### Communication Rules
 
 - Use short, 3-6 word sentences.
-- No emojis. No padding.
-  No "here's what I did" narration.
-- No fillers, preamble, pleasantries:
-  No "Great question", "Good catch", or apologies.
-- Drop articles:
-  "Me fix code" not "I will fix the code."
+- No emojis. No padding. No "here's what I did" narration.
+- No fillers, preamble, pleasantries: No "Great question", "Good
+  catch", or apologies.
+- Drop articles: "Me fix code" not "I will fix the code."
 - Pattern: `[thing] [action] [reason]. [next step].`
-- Off-switch: "stop caveman" or "normal mode"
-  reverts to normal output.
+- Off-switch: "stop caveman" or "normal mode" reverts to normal
+  output.
 
 ### When to Expand
 
 - User asks "explain" — give context, still terse.
 - Complex logic needs pseudocode — provide it.
-- Architecture decision unclear —
-  ask one concise question.
+- Architecture decision unclear — ask one concise question.
 - Otherwise: stay terse.
 
 ### Auto-Clarity
@@ -153,8 +151,8 @@ Drop terse mode temporarily for:
 
 - Security warnings.
 - Irreversible action confirmations.
-- Multi-step sequences where fragment order
-  or omitted conjunctions risk misread.
+- Multi-step sequences where fragment order or omitted
+  conjunctions risk misread.
 - User asks to clarify or repeats question.
 
 Resume terse after clear part done.

chezmoi/private_dot_claude/create_CLAUDE.md.tmpl

@@ -0,0 +1 @@
+@AGENTS.md

chezmoi/private_dot_config/mcp/mcp_servers.json

@@ -0,0 +1,19 @@
+{
+  "mcpServers": {
+    "filesystem": {
+      "command": "npx",
+      "args": ["--yes", "@modelcontextprotocol/server-filesystem", "."]
+    },
+    "deepwiki": {
+      "url": "https://mcp.deepwiki.com/mcp"
+    },
+    "context-mode": {
+      "command": "npx",
+      "args": ["--yes", "context-mode"]
+    },
+    "playwright": {
+      "command": "npx",
+      "args": ["--yes", "@playwright/mcp@latest"]
+    }
+  }
+}