GitHub Tool Catalog for AI Agents

A small, conservative, machine-readable catalog of GitHub-adjacent tools for AI agents and the humans supervising them.

This README is generated from catalog/tools.yml. Edit the catalog, then run python scripts/render_readme.py.

Start Here

Workflow	Use first	Add when	Caution
Agent repo maintenance	Agent repo maintenance kit	Browser validation kit for UI behavior	Limit tokens and require human approval for writes.
Human daily review	Human daily review kit	gh-dash when terminal queues help	Review local diffs before push or merge.
Polyrepo work	Human polyrepo work kit	ghq for local clone organization	ghq does not replace GitHub API tooling.
Lean or proof repo	Proof repo support kit	Context7 when docs are relevant	Bind proof tooling to the active project.
Agentic workflow evaluation	Evaluation kit	Only after team workflow review	Keep it out of default rollout until permissions, billing, and review boundaries are clear.

Starter Kits

Agent repo maintenance

Inspect GitHub state, navigate code, fetch current docs, and prepare supervised repository changes.

• Core: GitHub CLI, GitHub MCP Server, Serena, Context7
• Caution: Use limited GitHub scopes, project-scoped code tools, and human approval for writes.
• Guardrails: limited-token, read-only-mode, project-scope, human-approval
• Install:
  • brew install gh
  • codex mcp add github --url https://api.githubcopilot.com/mcp/ --bearer-token-env-var GITHUB_PAT_TOKEN
  • codex mcp add serena -- serena start-mcp-server --project-from-cwd --context=codex
  • codex mcp add context7 -- npx -y @upstash/context7-mcp@latest

Agent browser validation

Validate UI and browser behavior through constrained automation and diagnostics.

• Core: Playwright MCP
• Optional: Chrome DevTools MCP
• Caution: Prefer isolated browser contexts and non-production targets before using authenticated profiles.
• Guardrails: isolated-browser, no-user-profile, non-production
• Install:
  • codex mcp add playwright -- npx -y @playwright/mcp@latest
  • codex mcp add chrome-devtools -- npx -y chrome-devtools-mcp@latest --slim --headless --isolated

Human daily review

Review pull requests, inspect diffs, and manage local Git state during everyday maintenance.

• Core: GitHub CLI, lazygit, delta
• Optional: gh-dash
• Caution: Review local diffs before push or merge; gh-dash remains a reviewed extension, not a required default.
• Guardrails: human-approval
• Install:
  • brew install gh
  • brew install lazygit
  • brew install git-delta
  • gh extension install dlvhdr/gh-dash

Human polyrepo work

Keep many local repositories organized while preserving GitHub operations in gh.

• Core: GitHub CLI, lazygit, delta
• Optional: ghq
• Caution: Use ghq for local repository organization only; do not treat it as GitHub API coverage.
• Guardrails: project-scope, human-approval
• Install:
  • brew install gh
  • brew install lazygit
  • brew install git-delta
  • brew install ghq

Proof repo support

Support Lean proof work with scoped diagnostics, PR readback, and readable diffs.

• Core: lean-lsp-mcp, GitHub CLI, delta
• Optional: Context7
• Caution: Bind Lean tooling to the active proof project before exposing diagnostics to agents.
• Guardrails: project-scope
• Install:
  • codex mcp add lean-lsp --env LEAN_PROJECT_PATH=/path/to/lean/project -- uvx lean-lsp-mcp
  • brew install gh
  • brew install git-delta
  • codex mcp add context7 -- npx -y @upstash/context7-mcp@latest

Evaluation Kits

These are not default stacks. Use them only for reviewed experiments or team-specific workflows.

Agentic workflow evaluation

Evaluate stacked PRs, Actions-backed agent workflows, or a separate terminal agent runtime.

• Core: gh-stack, GitHub Agentic Workflows, GitHub Copilot CLI
• Caution: Evaluation only: use sandbox repositories, define billing and review boundaries, and avoid default rollout claims.
• Guardrails: sandbox-repo, human-approval, non-production
• Install:
  • gh extension install github/gh-stack
  • gh extension install github/gh-aw
  • brew install copilot-cli

Daily Workflows

• Agent PR review: inspect issue and PR state with GitHub MCP Server or GitHub CLI, navigate code with Serena, fetch current docs with Context7, validate behavior with Playwright MCP, then read back checks with GitHub CLI.
• Human PR review: use GitHub CLI for PR state, delta for readable diffs, lazygit for local Git actions, and gh-dash only when a terminal review queue helps.
• Browser validation: start with Playwright MCP; add Chrome DevTools MCP when DevTools-level network, console, performance, or screenshot diagnostics are needed.
• Lean or proof repository: bind project scope for lean-lsp-mcp, use GitHub CLI for PR/check readback, and use delta for proof diff review.
• Agentic workflow evaluation: keep gh-stack, GitHub Agentic Workflows, and GitHub Copilot CLI in sandbox or explicitly reviewed workflows before routine rollout.

Selection Criteria

• Helps AI agents or maintainers work with GitHub, pull requests, CI, browser testing, code intelligence, documentation, or proof workflows.
• Has a public source or documentation URL.
• Can be classified by usefulness, operational risk, recommendation status, and practical use constraints.
• Uses watch for promising tools that are preview-only, broad-permission, or not yet a safe default.

Recommendation Legend

Status	Meaning
Recommended	Good default for the right workflow.
Situational	Useful when the workflow clearly needs it.
Watch	Promising, preview-only, heavy, or not yet a safe default.
Avoid	Known mismatch for this catalog's default recommendations.

Recommended Defaults

Tool	Best for	Agent	Human	Risk	Caution
Chrome DevTools MCP	browser-debugging, performance, network-inspection	high	medium	High	Use slim, headless, isolated mode unless user profile state is explicitly required.
Context7	current-docs, framework-apis, agent-research	high	medium	Medium	Use it to support implementation claims, then verify behavior with project tests or official sources.
delta	diff-review, readability, git-pager	medium	high	Low	Configure as a pager first; it improves review readability without changing GitHub state.
GitHub CLI	github-api, pr-review, actions-readback, release-workflows	high	high	Medium	Authenticate with the minimum scopes needed for the current repository or organization task.
GitHub MCP Server	agent-github-api, issues-prs, workflow-readback	high	medium	High	Start with the narrowest available token scope or read-only toolset before allowing write operations.
lazygit	local-git, branch-review, staging	medium	high	Medium	Review staged changes before committing or pushing.
lean-lsp-mcp	lean-proof, proof-diagnostics, formal-methods	high	medium	Medium	Bind LEAN_PROJECT_PATH to the active Lean project before exposing proof tooling.
Playwright MCP	browser-validation, ui-testing, agent-e2e	high	medium	High	Run against a constrained browser context and non-production target when possible.
Serena	code-intelligence, repo-navigation, agent-editing	high	medium	High	Start with project-scoped startup from the active repository, not broad filesystem access.

Watch / Avoid Decisions

Tool	Status	Daily use	Risk	Why not a default
gh-news	Watch	Not default	Medium	Promising notification-focused gh extension; it uses token-backed GitHub notification actions, so keep it in watch until publisher trust and team fit are reviewed.
gh-stack	Watch	Not default	High	Preview-oriented stacked PR workflow; use only when stacked PRs are enabled and the team's auth model supports the extension.
GitHub Agentic Workflows	Watch	Not default	High	Powerful but operationally heavy; treat permissions, billing, guardrails, and workflow safety as first-class review items.
gitu	Watch	Not default	Medium	Promising Magit-style TUI with narrower adoption than lazygit and gitui; keep in watch until use cases are clearer.
hub	Avoid	Not default	Medium	Keep only to explain migration pressure toward gh; do not recommend for new GitHub CLI workflows.

Full Catalog

GitHub MCP / Connectors

GitHub MCP Server

Official GitHub MCP server for repository, issue, pull request, workflow, project, org, and gist access from AI agents.

• Repo: github/github-mcp-server
• Kind: mcp
• Status: recommended; risk: high; daily use: reviewed
• Best for: agent-github-api, issues-prs, workflow-readback
• Agent use: Let an agent inspect and update GitHub repository, issue, pull request, workflow, project, org, or gist state when the task explicitly needs GitHub context.
• Human use: Use through an agent when GitHub state needs exact readback or controlled repository actions.
• Cautious start: Start with the narrowest available token scope or read-only toolset before allowing write operations.
• Guardrails: limited-token, read-only-mode, human-approval
• Pairs with: GitHub CLI (high), Serena (high), Context7 (high)
• Agent usefulness: high
• Human usefulness: medium
• Install: codex mcp add github --url https://api.githubcopilot.com/mcp/ --bearer-token-env-var GITHUB_PAT_TOKEN
• Notes: Use targeted toolsets or read-only mode when broad write access is unnecessary.

GitHub CLI and API

GitHub CLI

Official GitHub command line interface for issues, pull requests, Actions, releases, rulesets, API calls, and authenticated automation.

• Repo: cli/cli
• Kind: cli
• Status: recommended; risk: medium; daily use: routine
• Best for: github-api, pr-review, actions-readback, release-workflows
• Agent use: Use for auditable GitHub operations, exact API calls, and PR/check readback when connector coverage is insufficient.
• Human use: Use as the baseline command line for issues, pull requests, Actions, releases, and authenticated automation.
• Cautious start: Authenticate with the minimum scopes needed for the current repository or organization task.
• Guardrails: limited-token, human-approval
• Pairs with: lazygit (medium), delta (medium), GitHub MCP Server (high)
• Agent usefulness: high
• Human usefulness: high
• Install: brew install gh
• Notes: Prefer gh api for exact REST or GraphQL operations that are not exposed by a connector or MCP tool; treat gh skill as a gh capability rather than a separate tool entry.

gh Extensions

gh-dash

Terminal dashboard for GitHub pull requests, issues, custom sections, custom actions, and keyboard- driven triage.

• Repo: dlvhdr/gh-dash
• Kind: gh-extension
• Status: situational; risk: medium; daily use: reviewed
• Best for: human-triage, pr-queues, issue-queues
• Agent use: Low fit for autonomous agents; treat as a human-facing queue dashboard.
• Human use: Use for keyboard-driven PR and issue triage when a terminal dashboard improves review flow.
• Cautious start: Install only after reviewing third-party extension trust and terminal UI assumptions for the team.
• Guardrails: human-approval
• Pairs with: GitHub CLI (medium), lazygit (medium)
• Agent usefulness: low
• Human usefulness: high
• Install: gh extension install dlvhdr/gh-dash
• Notes: Great for human review queues; audit third-party extension trust and Nerd Font/UI assumptions before default rollout.

gh-news

GitHub CLI notification TUI for reading, filtering, previewing, muting, snoozing, and archiving GitHub notifications from the terminal.

• Repo: chmouel/gh-news
• Kind: gh-extension
• Status: watch; risk: medium; daily use: not_default
• Best for: notifications, human-triage, team-evaluation
• Agent use: Not an agent default; notifications usually need human prioritization.
• Human use: Evaluate for terminal notification review when the team accepts the extension trust model.
• Cautious start: Review publisher trust and token-backed notification actions before routine use.
• Guardrails: limited-token, human-approval
• Pairs with: GitHub CLI (medium)
• Agent usefulness: low
• Human usefulness: medium
• Install: gh extension install chmouel/gh-news
• Notes: Promising notification-focused gh extension; it uses token-backed GitHub notification actions, so keep it in watch until publisher trust and team fit are reviewed.

gh-stack

GitHub extension for stacked branches and stacked pull requests, including agent-facing skill guidance.

• Repo: github/gh-stack
• Kind: gh-extension
• Status: watch; risk: high; daily use: not_default
• Best for: stacked-prs, branch-stacks, team-evaluation
• Agent use: Evaluate only for teams that explicitly run stacked PR workflows with review boundaries.
• Human use: Use when a team has adopted stacked pull requests and understands the extension workflow.
• Cautious start: Try in a sandbox repository before using it on production branches.
• Guardrails: sandbox-repo, human-approval
• Pairs with: GitHub CLI (high)
• Agent usefulness: medium
• Human usefulness: high
• Install: gh extension install github/gh-stack
• Notes: Preview-oriented stacked PR workflow; use only when stacked PRs are enabled and the team's auth model supports the extension.

Agentic Workflows

GitHub Copilot CLI

GitHub Copilot coding agent for terminal workflows, with configurable permissions and built-in GitHub MCP support.

• Repo: github/copilot-cli
• Kind: agent-cli
• Status: situational; risk: high; daily use: reviewed
• Best for: terminal-agent, agent-runtime, human-supervised-coding
• Agent use: Use as a separate terminal-native agent runtime when the workflow calls for Copilot CLI rather than Codex tooling.
• Human use: Use for human-supervised terminal agent tasks with configured permissions.
• Cautious start: Review permissions and repository write boundaries before using it on important branches.
• Guardrails: human-approval, project-scope
• Pairs with: GitHub CLI (high), Context7 (high)
• Agent usefulness: medium
• Human usefulness: high
• Install: brew install copilot-cli
• Notes: A separate agent runtime rather than a Codex tool; prefer this standalone CLI over retired legacy gh-copilot workflows.

GitHub Agentic Workflows

GitHub CLI extension for defining and running repository agent workflows through GitHub Actions.

• Repo: github/gh-aw
• Kind: agent-workflow
• Status: watch; risk: high; daily use: not_default
• Best for: agentic-workflows, github-actions, team-evaluation
• Agent use: Evaluate for repository agent workflows only with explicit billing, permission, and review boundaries.
• Human use: Use to prototype GitHub Actions-backed agent workflows after operational review.
• Cautious start: Start in a sandbox repository with clear workflow permissions and cost boundaries.
• Guardrails: sandbox-repo, human-approval, non-production
• Pairs with: GitHub CLI (high), GitHub Copilot CLI (high)
• Agent usefulness: medium
• Human usefulness: medium
• Install: gh extension install github/gh-aw
• Notes: Powerful but operationally heavy; treat permissions, billing, guardrails, and workflow safety as first-class review items.

Browser and Testing MCP

Chrome DevTools MCP

Chrome DevTools MCP server for navigation, script execution, screenshots, performance, network, and debugging workflows.

• Repo: ChromeDevTools/chrome-devtools-mcp
• Kind: mcp
• Status: recommended; risk: high; daily use: reviewed
• Best for: browser-debugging, performance, network-inspection
• Agent use: Let an agent inspect navigation, screenshots, performance, network, and console behavior in a constrained Chrome context.
• Human use: Use for browser debugging when DevTools-level inspection is more useful than a test runner.
• Cautious start: Use slim, headless, isolated mode unless user profile state is explicitly required.
• Guardrails: isolated-browser, no-user-profile, non-production
• Pairs with: Playwright MCP (high), GitHub CLI (high)
• Agent usefulness: high
• Human usefulness: medium
• Install: codex mcp add chrome-devtools -- npx -y chrome-devtools-mcp@latest --slim --headless --isolated
• Notes: The slim isolated mode is a safer default for general coding-agent use.

Playwright MCP

Browser automation MCP server for testing and data extraction through Playwright's accessibility- oriented model.

• Repo: microsoft/playwright-mcp
• Kind: mcp
• Status: recommended; risk: high; daily use: reviewed
• Best for: browser-validation, ui-testing, agent-e2e
• Agent use: Let an agent validate web behavior through Playwright accessibility and browser automation flows.
• Human use: Use when browser validation needs reproducible automation rather than manual clicking.
• Cautious start: Run against a constrained browser context and non-production target when possible.
• Guardrails: isolated-browser, no-user-profile, non-production
• Pairs with: GitHub CLI (high), Chrome DevTools MCP (high)
• Agent usefulness: high
• Human usefulness: medium
• Install: codex mcp add playwright -- npx -y @playwright/mcp@latest
• Notes: Prefer constrained browser contexts for agent runs that do not need local user profile state.

Code Intelligence MCP

Serena

Semantic code retrieval and editing MCP server for coding agents, backed by language-server style project understanding.

• Repo: oraios/serena
• Kind: mcp
• Status: recommended; risk: high; daily use: reviewed
• Best for: code-intelligence, repo-navigation, agent-editing
• Agent use: Let an agent navigate and edit a repository with project-scoped semantic code understanding.
• Human use: Use through an agent when semantic navigation is faster than plain text search.
• Cautious start: Start with project-scoped startup from the active repository, not broad filesystem access.
• Guardrails: project-scope, human-approval
• Pairs with: GitHub CLI (high), Context7 (high)
• Agent usefulness: high
• Human usefulness: medium
• Install: codex mcp add serena -- serena start-mcp-server --project-from-cwd --context=codex
• Notes: Best used with project-scoped startup rather than broad filesystem access.

Docs and Knowledge MCP

Context7

Documentation MCP server that injects current library docs and examples into agent prompts.

• Repo: upstash/context7
• Kind: mcp
• Status: recommended; risk: medium; daily use: routine
• Best for: current-docs, framework-apis, agent-research
• Agent use: Fetch current library docs and examples before implementing against APIs likely to change.
• Human use: Use as an agent-assisted documentation source when current API behavior matters.
• Cautious start: Use it to support implementation claims, then verify behavior with project tests or official sources.
• Guardrails: human-approval
• Pairs with: Serena (medium), GitHub CLI (medium)
• Agent usefulness: high
• Human usefulness: medium
• Install: codex mcp add context7 -- npx -y @upstash/context7-mcp@latest
• Notes: Useful when framework or API behavior is likely to have changed since model training.

Formal Proof / Lean support

lean-lsp-mcp

Lean LSP MCP server for proof diagnostics, goal inspection, Loogle-style search, and optional Lean REPL acceleration.

• Repo: oOo0oOo/lean-lsp-mcp
• Kind: mcp
• Status: recommended; risk: medium; daily use: specialist
• Best for: lean-proof, proof-diagnostics, formal-methods
• Agent use: Let an agent inspect Lean proof goals, diagnostics, and search in a bound Lean project.
• Human use: Use for proof development support when working in Lean repositories.
• Cautious start: Bind LEAN_PROJECT_PATH to the active Lean project before exposing proof tooling.
• Guardrails: project-scope
• Pairs with: GitHub CLI (medium), delta (medium)
• Agent usefulness: high
• Human usefulness: medium
• Install: codex mcp add lean-lsp --env LEAN_PROJECT_PATH=/path/to/lean/project -- uvx lean-lsp-mcp
• Notes: Strong fit for Lean repositories; scope LEAN_PROJECT_PATH to the active proof project.

Git TUI

lazygit

Terminal UI for everyday Git workflows such as staging, committing, rebasing, stashing, branch navigation, and conflict-oriented local review.

• Repo: jesseduffield/lazygit
• Kind: git-tui
• Status: recommended; risk: medium; daily use: routine
• Best for: local-git, branch-review, staging
• Agent use: Medium fit; useful mainly when a human supervises local Git state around agent changes.
• Human use: Use for everyday staging, commits, rebases, stashes, branch navigation, and conflict review.
• Cautious start: Review staged changes before committing or pushing.
• Guardrails: human-approval
• Pairs with: GitHub CLI (medium), delta (medium)
• Agent usefulness: medium
• Human usefulness: high
• Install: brew install lazygit
• Notes: Best default local Git TUI; pair with gh or gh-dash for GitHub pull request and issue workflows.

gitui

Fast Rust terminal UI for local Git workflows, including diff, commit, stash, blame, log, and index operations.

• Repo: gitui-org/gitui
• Kind: git-tui
• Status: situational; risk: medium; daily use: specialist
• Best for: local-git, rust-tui, terminal-review
• Agent use: Low fit for agents; use mainly as a human local Git interface.
• Human use: Use as a lightweight Rust TUI alternative for local Git operations.
• Cautious start: Keep it as an alternative until it clearly fits the user workflow better than lazygit.
• Guardrails: human-approval
• Pairs with: GitHub CLI (medium), delta (medium)
• Agent usefulness: low
• Human usefulness: medium
• Install: brew install gitui
• Notes: Lightweight Rust Git TUI alternative; keep as an alternative rather than the default over lazygit.

tig

Mature text-mode interface for browsing Git history, diffs, refs, blame, and repository state in minimal terminal environments.

• Repo: jonas/tig
• Kind: git-tui
• Status: situational; risk: low; daily use: specialist
• Best for: git-history, ssh-terminal, minimal-environments
• Agent use: Low fit for agents; use as a simple history and diff browser in constrained terminals.
• Human use: Use for browsing history, blame, refs, and diffs in minimal terminal environments.
• Cautious start: Use for read-heavy inspection before changing local Git state.
• Guardrails: human-approval
• Pairs with: GitHub CLI (low), delta (low)
• Agent usefulness: low
• Human usefulness: medium
• Install: brew install tig
• Notes: Good fallback for SSH, server, and low-dependency environments; less guided than newer Git TUIs.

gitu

Magit-inspired terminal Git client for hunk and line staging, commits, fixups, rebases, stashes, and local history workflows.

• Repo: altsem/gitu
• Kind: git-tui
• Status: watch; risk: medium; daily use: not_default
• Best for: magit-style-git, hunk-staging, team-evaluation
• Agent use: Not an agent default; evaluate only for local Git workflows that need Magit-style interaction.
• Human use: Evaluate for hunk and line staging when the team wants a Magit-inspired terminal client.
• Cautious start: Try on a non-critical branch before adopting it as a daily Git interface.
• Guardrails: human-approval, sandbox-repo
• Pairs with: delta (medium)
• Agent usefulness: low
• Human usefulness: medium
• Install: cargo install gitu
• Notes: Promising Magit-style TUI with narrower adoption than lazygit and gitui; keep in watch until use cases are clearer.

Git Diff and Repo Utilities

delta

Syntax-highlighting pager for Git, diff, grep, ripgrep JSON, and blame output that improves review readability.

• Repo: dandavison/delta
• Kind: git-pager
• Status: recommended; risk: low; daily use: routine
• Best for: diff-review, readability, git-pager
• Agent use: Use to make diff and review output easier for humans supervising agent changes.
• Human use: Use as a pager upgrade for readable Git, diff, grep, ripgrep JSON, and blame output.
• Cautious start: Configure as a pager first; it improves review readability without changing GitHub state.
• Pairs with: GitHub CLI (medium), lazygit (medium)
• Agent usefulness: medium
• Human usefulness: high
• Install: brew install git-delta
• Notes: High-value diff and pager upgrade; improves review readability without replacing Git or GitHub workflows.

ghq

Local repository manager that standardizes clone locations and makes multi-host or polyrepo workspaces easier to navigate.

• Repo: x-motemen/ghq
• Kind: repo-manager
• Status: situational; risk: low; daily use: reviewed
• Best for: repo-organization, polyrepo, local-workspaces
• Agent use: Medium fit for agents that need predictable local clone locations across many repositories.
• Human use: Use to standardize clone locations and navigate multi-host or polyrepo workspaces.
• Cautious start: Treat it as local repository organization, not as GitHub API or PR tooling.
• Guardrails: project-scope
• Pairs with: GitHub CLI (low), lazygit (medium)
• Agent usefulness: medium
• Human usefulness: high
• Install: brew install ghq
• Notes: Useful for polyrepo and multi-host local repository organization; not a GitHub API tool.

Legacy / Avoid

hub

Legacy command line wrapper that extends Git with GitHub operations but has largely been superseded by the official GitHub CLI.

• Repo: mislav/hub
• Kind: cli
• Status: avoid; risk: medium; daily use: not_default
• Best for: legacy-migration, non-recommendation
• Agent use: Do not use for new agent workflows; prefer GitHub CLI.
• Human use: Keep only as migration context for legacy users moving to gh.
• Cautious start: Do not start new GitHub workflows with hub; migrate to gh instead.
• Guardrails: human-approval
• Agent usefulness: low
• Human usefulness: low
• Install: brew install hub
• Notes: Keep only to explain migration pressure toward gh; do not recommend for new GitHub CLI workflows.

Machine-Readable Data

• Source of truth: catalog/tools.yml
• Schema: schema/tool.schema.json
• Generated JSON: dist/catalog.v2.json
• Generated README: README.md

Governance and Contributions

• See CONTRIBUTING.md before opening issues or pull requests.
• See docs/curation-governance.md for inclusion, removal, taxonomy, evidence, and recommendation rules.
• See docs/maintenance.md for refresh cadence, automation policy, and public roadmap rules.
• See docs/pr-review-checklist.md for maintainer review expectations.

Validation

python scripts/validate_catalog.py
python scripts/render_readme.py --check
python scripts/render_catalog_json.py --check
python scripts/check_links.py

License

CC0-1.0. See LICENSE.