PAL MCP Server: Setup for Claude Code, Cursor and Other Apps

• Brainstorms with another AI model and returns its answer

  Starts a conversation with the model you name, such as Gemini or an OpenAI model, so you get a second opinion inside your current session.

  Try: “Ask Gemini whether Redis or Memcached fits this cache better.”

  Network · chat

• Has another model reason through a hard problem at length

  Runs a step-by-step deep reasoning pass with extra checks on edge cases and alternatives, then asks a second model to review it.

  Try: “Think through the edge cases of this payment retry logic with Gemini.”

  Network · thinkdeep

• Breaks a big project into a step-by-step plan

  Builds a structured plan one step at a time, and can revise earlier steps as the plan develops.

  Try: “Plan the migration from REST to GraphQL in small steps.”

  Read · planner

• Asks several AI models for their opinion and compares them

  Collects views from more than one model on a proposal, and lets you steer each model to argue for or against.

  Try: “Get consensus from Gemini and o3 on switching to a monorepo.”

  Network · consensus

• Reviews code and lists issues by severity

  Walks through the code step by step and reports problems ranked by severity, with a second model checking the result.

  Try: “Review the auth module with Gemini Pro and rank the problems.”

  Files · codereview

• Checks your changes before you commit them

  Looks over the pending changes in a git repo for bugs and regressions, with a second model validating the findings.

  Try: “Run a pre-commit check on my staged changes.”

  Files · precommit

• Investigates a bug and works toward the root cause

  Guides a step-by-step investigation of an error, tracking hypotheses and evidence, then asks a second model to confirm.

  Try: “Debug this stack trace with o3 and find the root cause.”

  Files · debug

• Hands a task to another AI command line tool such as Gemini CLI or Codex

  Launches an external AI CLI as a sub-agent with a role such as planner or code reviewer, and returns its answer. The docs say it starts those tools with relaxed permission flags so they can edit files and run tools on their own.

  Try: “Use clink with codex codereviewer to audit the auth module.”

  Runs code · clink

• Makes the assistant question a claim instead of agreeing

  Wraps a statement in a prompt that asks for critical analysis, so the assistant does not just say you are right.

  Try: “Challenge this claim before we build on it.”

  Read · challenge

• Reminds the assistant to look up current docs for an API or library

  Returns instructions that tell the assistant to research up-to-date documentation in a separate sub-task. It has no web search of its own.

  Try: “Look up the current docs for the Stripe Checkout API.”

  Read · apilookup

• Lists the AI models you can use with your keys

  Shows the models available from each provider you have configured. It cannot be switched off.

  Try: “Which models can PAL use right now?”

  Read · listmodels

• Shows the server version and system information

  Reports the PAL version and details about the machine it runs on. It cannot be switched off.

  Try: “What version of PAL is running?”

  Read · version

• Studies a codebase for structure, patterns and dependencies

  Examines files and reports on architecture, patterns and dependencies. Off by default in the README setup.

  Try: “Analyze the architecture of this repo.”

  Files · analyze

• Finds refactoring opportunities and plans them

  Analyses code step by step and suggests how to break up large pieces. Off by default in the README setup.

  Try: “Find refactors that would split up this 2,000 line file.”

  Files · refactor

• Plans tests, including edge cases

  Works out a test plan for code, with edge cases, validated by a second model. Off by default in the README setup.

  Try: “Generate tests for the invoice calculator.”

  Files · testgen

• Audits code for security problems

  Runs a security review against the OWASP Top 10 and compliance areas. Off by default in the README setup.

  Try: “Run a security audit on the login flow.”

  Files · secaudit

• Plans documentation for code

  Works through code to produce documentation guidance, including how complex each part is. Off by default in the README setup.

  Try: “Document the public functions in this module.”

  Files · docgen

• Maps how a function is called through the code

  Produces static-analysis prompts for call-flow mapping. Off by default in the README setup.

  Try: “Trace everything that calls the checkout function.”

  Files · tracer

Watch out for

• clink launches other AI tools with relaxed permissions

  The docs say Gemini starts with its auto-approve flag, Codex with sandbox checks bypassed and Claude with edits accepted. Turn clink off in the disabled-tools setting if you do not need it. github.com

• Your code is sent to third-party model providers

  Files and prompts go to whichever provider handles the request. Use a local Ollama model for code that must stay on your machine. github.com

• Several tools are on by default

  The code enables all 18 tools unless you set a disabled list. The README suggests turning six off to save context. github.com

• Multi-step tools can run up provider costs

  Review and debug workflows make several model calls. Set spending limits on the keys you give it. github.com

Security advisories