CLI Coming Soon
    Back to catalog

    Claude Code settings.json validator

    Published · Last updated · Docs checked 2026-11-12

    Claude Code reads settings.json from ~/.claude/settings.json, .claude/settings.json and .claude/settings.local.json. Each is strict JSON. Permission rules look like Tool or Tool(specifier), env holds string values, and hooks nest event, matcher group, then handler. Paste yours below to check it.

    Validate your settings.json

    Runs in your browser. Nothing you paste or drop is uploaded or stored. No AI model reads it.

    Results show here as soon as there is something to check.

    A Claude Code settings.json example

    This is the example on Anthropic's settings page, copied as written. It lets Claude Code run your lint and test commands without asking and stops it reading .env files. It passes every check here.

    {
      "$schema": "https://json.schemastore.org/claude-code-settings.json",
      "permissions": {
        "allow": [
          "Bash(npm run lint)",
          "Bash(npm run test *)"
        ],
        "deny": [
          "Read(./.env)",
          "Read(./.env.*)"
        ]
      }
    }
    

    The $schema line gives you autocomplete in editors that read JSON schema. For a personal file, a team file and an organization file with every key explained, see Anthropic's example settings files.

    Where is settings.json?

    From Settings files and who they affect:

    ScopeFileWho it affects
    User~/.claude/settings.jsonYou, in every project on this machine
    Shared project.claude/settings.jsonEveryone working in the folder that contains it. Commit it in git.
    Project local.claude/settings.local.jsonYou, in this one project only. Keep it out of git.
    Managedmanaged-settings.json and other managed sourcesEveryone your organization deploys it to

    A fifth file, ~/.claude.json, holds your account, MCP server configs and a short list of global config keys. Claude Code writes it itself. When the same key is in more than one file, the settings precedence rules pick the value.

    Permissions, env and hooks: the shapes the docs give

    Permissions

    "Permission rules follow the format Tool or Tool(specifier)" (Permission rule syntax). A trailing * matches the rest of a command. Put the * after the subcommand: Bash(git log *) allows only git log, while Bash(git *) allows every git command. MCP tools are named mcp__server__tool, and an MCP rule with parentheses is skipped.

    {
      "permissions": {
        "allow": ["Bash(npm run *)"],
        "ask": ["Bash(git push *)"],
        "deny": ["Read(./.env)"],
        "defaultMode": "acceptEdits"
      }
    }

    env

    An object that maps variable names to string values (env). Write "1", not 1.

    {
      "env": {
        "DISABLE_AUTO_COMPACT": "1",
        "ANTHROPIC_BASE_URL": "https://proxy.example.com"
      }
    }

    Hooks

    Three levels: a hook event such as PreToolUse, a matcher group that filters it, and one or more handlers with a type of command, http, mcp_tool, prompt or agent (Hooks: Configuration). The docs list 33 events as of 2026-11-12.

    {
      "hooks": {
        "PreToolUse": [
          {
            "matcher": "Bash",
            "hooks": [
              { "type": "command", "command": "~/.claude/hooks/check-bash.sh" }
            ]
          }
        ]
      }
    }

    The env, permissions and hooks examples are copied from the settings reference.

    What it checks

    A problem means Claude Code rejects the file or the value has the wrong type. A warning means Claude Code skips, ignores or never matches the entry, or the docs do not list it. A tip is worth a look. A key the docs do not list is reported as "not in the docs as of 2026-11-12", since it may be newer than the docs.

    CheckWhat the docs say
    Strict JSON, with the line and column of the first error"Settings files are strict JSON: a // comment or a trailing comma is a syntax error, and Claude Code reports the file as a Settings Error at the next start." Settings: Edit a settings file
    The same key twice in one object"The names within an object SHOULD be unique." RFC 8259, section 4 (Objects)
    Top-level keys the settings reference does not list"This reference page lists each key Claude Code reads from a settings file, plus the short group of keys it keeps in ~/.claude.json instead." All settings: Settings index
    Keys that belong in ~/.claude.json instead"Save these keys in ~/.claude.json, not in a settings file. Claude Code ignores them anywhere else." All settings: Global config settings
    Keys set in a file the docs say ignores them"Scope lists the files it can go in: User is ~/.claude/settings.json, Project is .claude/settings.json, Local is .claude/settings.local.json, and Managed is what your organization deploys." All settings: Settings index
    Keys the settings index marks as deprecated"This reference page lists each key Claude Code reads from a settings file, plus the short group of keys it keeps in ~/.claude.json instead." All settings: Settings index
    permissions is an object with documented keys and string arrays"Type: object with allow, ask, deny, additionalDirectories, blockReadsOutsideWorkingDirectories, blockWritesOutsideWorkingDirectories, defaultMode, denyInsteadOfPrompting, disableBypassPermissionsMode, and disableAutoMode" All settings: permissions
    Each rule is Tool or Tool(specifier)"Permission rules follow the format Tool or Tool(specifier). Parentheses inside the specifier are literal, so a command or path that contains them needs no escaping." Permissions: Permission rule syntax
    Tool names match the tools reference"A deny or ask rule whose tool name matches no known tool produces a startup warning to catch typos. Tool names containing _ or * are exempt from the check, and so are the names of tools Claude Code has removed, such as TaskOutput." Permissions: Tool name wildcards
    Rules Claude Code skips or never matches"An unanchored allow glob such as "*", "B*", or "mcp__*" is skipped and doesn't auto-approve anything." Permissions: Tool name wildcards
    The same rule in allow and in deny or ask"Claude Code evaluates deny rules first, then ask, then allow, and the first match decides regardless of how specific each rule is" All settings: Permission rule syntax
    permissions.defaultMode is a documented mode"Type: string, one of: "default", "acceptEdits", "plan", "auto", "dontAsk", "bypassPermissions", "manual"" All settings: permissions.defaultMode
    env is an object of string values"Type: object mapping variable names to string values" All settings: env
    Hook event names match the hooks reference"Settings Warning: only individual entries fail, such as a malformed permission rule or an unknown hook event name. Claude Code skips those values and keeps the rest of the file in effect." Hooks: Hook lifecycle
    Each event holds matcher groups with a hooks array"Type: object keyed by hook event; each value is an array of { "matcher", "hooks" } groups whose hooks entries have a type of "command", "prompt", "agent", "http", or "mcp_tool"" All settings: hooks
    Each handler has a known type and its required fields"type | yes | "command", "http", "mcp_tool", "prompt", or "agent"" Hooks: Common fields
    model is a string, and not a near miss of an alias"Either of these warnings means Claude Code doesn't recognize the model ID you set, and the ID nearly spells a model name or alias it does recognize" Model configuration: Fix a mistyped model name

    How it works

    1. A small JSON parser reads the text and keeps the position of every key and value. On a syntax error it stops and names the line and column, with a hint for the common cases: a // comment, a trailing comma, a missing comma, single quotes, an unquoted key.
    2. It compares each top-level key with the 186 keys in the settings index and their scope column. If you say which file you pasted, it also flags keys the index says that file ignores.
    3. Each permission rule is split into a tool name and an optional specifier. The tool name is matched against the 46 tools in the tools reference, and the rule against the cases the permissions page says Claude Code skips or never matches.
    4. env, hooks and model are checked against their entries in the settings reference, the hooks reference and the model aliases table.
    5. The key list, hook events, tool names and aliases were read from the docs' Markdown on 2026-11-12 by a script in the site's repo. The sha256 of each page is listed under Sources, so you can tell whether a page has changed since.

    What it does not check

    • The values of most keys, such as sandbox, statusLine or attribution. Open the key in the settings reference for its type.
    • Whether your scripts, paths and MCP servers exist. It sees only the text you paste.
    • Full model IDs. Any ID that starts with claude- passes, and gateway or cloud IDs are not checked.
    • How files combine. Run /status to see which files loaded and claude doctor for the entries Claude Code dropped.

    All 186 settings keys in the docs

    Every top-level key in the settings index as of 2026-11-12, with its scope. Dotted keys such as permissions.allow nest under their parent.

    Show the key list
    KeyWhat it doesScope
    advisorModelPick which model answers when Claude asks the advisor toolAny file
    agentStart every session as a named subagent with its prompt, tools, and modelAny file
    agentPushNotifEnabledLet Claude send a push notification to your phone when it decides toAny file
    allowAllClaudeAiMcpsLoad the claude.ai connectors Claude Code fetches itself alongside a deployed managed-mcp.jsonManaged
    allowClaudeInChromeWithManagedMcpLet the built-in Claude in Chrome server run alongside a deployed managed-mcp.jsonManaged
    allowedChannelPluginsReplace the default allowlist of channel plugins that can push messagesManaged
    allowedHttpHookUrlsLimit which URLs HTTP hooks can targetAny file
    allowedMcpServersAllowlist which MCP servers users can addAny file
    allowedProvidersLimit which API providers a machine may useManaged
    allowManagedHooksOnlyRun only the hooks your organization deploysManaged
    allowManagedMcpServersOnlyMake the managed MCP allowlist the only one that appliesManaged
    allowManagedPermissionRulesOnlyMake managed settings the only settings source of permission rulesManaged
    alwaysThinkingEnabledTurn extended thinking off for every sessionAny file
    apiKeyHelperGenerate the API credential with your own commandAny file
    askUserQuestionTimeoutLet an unanswered question auto-continue after idle timeUser or managed
    appendPluginsRun your organization's mods after every mod a user installsUser or managed
    attributionCustomize the attribution Claude Code adds to commits and pull requestsAny file
    autoCompactEnabledTurn automatic compaction off or onAny file
    autoCompactWindowSet how full the context gets before Claude Code compactsAny file
    autoConnectIdeConnect to a running VS Code or JetBrains IDE automatically from an external terminalGlobal config
    autoContinueAtUsageLimitWait in the open session and continue the task automatically after a claude.ai usage limit resetsUser or managed
    autoInstallIdeExtensionLet /ide install the IDE extension from an external terminalGlobal config
    autoMemoryDirectoryStore auto memory in a directory you chooseAny file
    autoMemoryEnabledTurn auto memory off or onAny file
    autoModeAdd your own allow and deny rules to the auto mode classifierUser or managed
    autoScrollEnabledFollow new output to the bottom in fullscreen renderingAny file
    autoUpdatesChannelFollow the stable release channel instead of latestAny file
    availableModelsRestrict which models people can pickAny file
    availableModelsMatchMake each availableModels model ID entry permit only the version it namesManaged
    awaySummaryEnabledTurn off the session recap shown when you come back to the terminalAny file
    awsAuthRefreshRefresh expired Bedrock credentials in .aws with your own commandAny file
    awsCredentialExportSupply Bedrock credentials as JSON from your own commandAny file
    axScreenReaderRender screen-reader friendly outputUser or managed
    bashEditDiffEnabledRecord the files that changed while a Bash command ran in every permission modeUser or managed
    bashOutputMaxCharsSet how much of a successful command's output Claude receives inlineAny file
    blockedMarketplacesBlock plugin marketplace sources for your organizationManaged
    browserExternalPageToolsKeep Claude's tools off external pages in the desktop Browser paneManaged
    channelsEnabledAllow channels for your organizationManaged
    claudeInChromeDefaultEnabledTurn on Chrome integration when a session starts, in the interactive CLI and the VS Code extensionGlobal config
    claudeMdInject organization-wide CLAUDE.md instructions from managed settingsManaged
    claudeMdExcludesSkip specific CLAUDE.md files when memory loadsAny file
    cleanupPeriodDaysChoose how many days Claude Code keeps transcripts before deleting themAny file
    companyAnnouncementsShow your organization's announcements at startupAny file
    copyFullResponseMake /copy copy the full response without showing the pickerGlobal config
    copyOnSelectTurn off automatic copying of text you select with the mouse in fullscreen rendering and agent viewGlobal config
    crossSessionInboundChoose whether Claude Code delivers messages from your other sessions, shows a notice without delivering them, or refuses themAny file
    defaultShellChoose whether Bash or PowerShell runs the shell commands you type with the ! prefixAny file
    defaultToAgentsViewOpen agent view instead of a new conversation when you run claude with no argumentsGlobal config
    deniedMcpServersBlock specific MCP servers by URL, command, or nameAny file
    deniedModelsBlock specific models, even ones availableModels permitsManaged
    desktopSessionCleanupPeriodDaysSet an age limit in days for Claude Desktop and Cowork transcriptsUser or managed
    dialogExpirySet how long Claude Code waits for Remote Control or an SDK host to answer a forwarded dialog before it cancels the dialogUser or managed
    diffToolChoose whether Claude's proposed file changes open in the VS Code or JetBrains diff viewer or stay in the terminalGlobal config
    disableAgentViewTurn off background agents and agent viewAny file
    disableAllHooksTurn off hooks, a custom status line, and a custom @ file suggestion command at onceAny file
    disableArtifactDeprecated; use enableArtifact to turn the Artifact tool offAny file
    disableAutoModeRemove auto mode from the permission mode cycleAny file
    disableBrowserExternalNavigationLimit the desktop Browser pane to localhost for people and ClaudeManaged
    disableBundledSkillsTurn off the skills and workflows included with Claude CodeAny file
    disableClaudeAiConnectorsTurn off claude.ai connectors so Claude Code doesn't fetch themAny file
    disableCommandPluginSourcesBlock plugins that install by running a marketplace-declared commandManaged
    disableDeepLinkRegistrationStop Claude Code from registering the claude-cli:// handlerAny file
    disableDesktopLocalSessionsTurn off Desktop Code sessions that run on the device, leaving SSH to other hosts and cloudManaged
    disabledMcpjsonServersReject specific servers from a project's .mcp.jsonAny file
    disableMobileSimulatorToolsBlock Claude's tools in the desktop iOS Simulator paneManaged
    disableRemoteControlTurn off Remote Control everywhere it can startAny file
    disableRemotePreviewScriptsTurn off scripts in the desktop app's HTML previews in SSH and WSL sessionsManaged
    disableSideloadFlagsReject the CLI flags that sideload plugins, subagents, and MCP serversManaged
    disableSkillShellExecutionStop skills and custom commands from running inline shellAny file
    disableWorkflowsTurn dynamic workflows off for everyone; use enableWorkflows for yourselfAny file
    editorModeUse vim key bindings in the input promptAny file
    effortLevelSet a default effort level for models without a saved level of their ownAny file
    emojiCompletionEnabledTurn off :shortcode: emoji suggestions and replacement in the prompt inputAny file
    enableAllProjectMcpServersApprove every server in project .mcp.json files without a promptAny file
    enableArtifactTurn the Artifact tool off with a false in any file; no file can turn it back onAny file
    enabledMcpjsonServersApprove specific servers from a project's .mcp.jsonAny file
    enabledPluginsTurn individual plugins on or off per scopeAny file
    enableWorkflowsTurn dynamic workflows on or off against your plan's defaultAny file
    enforceAvailableModelsKeep the /model Default choice inside your availableModels allowlistAny file
    envSet environment variables for every session and its subprocessesAny file
    externalEditorContextShow Claude's last response as comments when you press Ctrl+G to editGlobal config
    extraKnownMarketplacesRegister marketplaces for a repository or an organizationAny file
    fallbackModelName backup models for when the primary is overloadedAny file
    fastModeTurn fast mode on for sessions where it's availableAny file
    fastModePerSessionOptInRequire people to turn fast mode on each sessionAny file
    feedbackDraftsControl whether Claude queues feedback drafts for you to reviewUser or managed
    feedbackSurveyRateChange how often the session quality survey appearsAny file
    fileCheckpointingEnabledTurn off or on the file snapshots that /rewind restoresAny file
    fileSuggestionSupply @ file autocomplete from your own commandAny file
    footerLinksRegexesMake issue or review IDs in output into clickable links below the input boxUser or managed
    forceLoginGatewayUrlSet the gateway URL the login screen connects toUser or managed
    forceLoginMethodRestrict login to claude.ai, Claude Console, or a cloud gatewayAny file
    forceLoginOrgUUIDPin claude.ai logins to your organization; only managed settings enforce itAny file
    forceRemoteSettingsRefreshBlock startup until server-managed settings are freshly fetchedManaged
    gatewayInternalNetworksLet /login reach a cloud gateway on public IPv4 space your organization uses internallyManaged
    gcpAuthRefreshRefresh Google Cloud credentials with your own commandAny file
    hooksRun your own commands as hooks at points in Claude Code's lifecycleAny file
    httpHookAllowedEnvVarsLimit which env vars HTTP hooks can put in headersAny file
    idleCompactionStop Claude Code from compacting a long conversation that you've stepped away fromAny file
    includeCoAuthoredByDeprecated; use attribution to hide or change commit and PR attributionAny file
    includeGitInstructionsRemove the built-in commit and PR instructions from Claude's contextAny file
    inputNeededNotifEnabledGet a push notification when Claude is waiting on youAny file
    isolatePeerMachinesAsk you before Claude messages one of your sessions on another machineAny file
    keybindingFlavorDeprecated and has no effect; the word-editing shortcuts always follow readline conventionsAny file
    languageHave Claude respond in a language other than EnglishAny file
    leftArrowOpensAgentsTurn off the ← shortcut that backgrounds the session and opens agent viewGlobal config
    managedMcpServersProvide remote MCP servers to every user alongside the ones they addManaged
    managedSourcesBehaviorCompose every managed source you deploy instead of using the highest-priority one aloneManaged
    maxEffortLevelCap the effort level for every model or per model, on every providerAny file
    maxProseWidthCap how wide the prose in Claude's responses runs in a wide terminalAny file
    minimumVersionKeep auto-updates from installing anything below a versionAny file
    modelChange the model Claude Code starts withAny file
    modelOverridesMap model IDs to your provider's IDs, such as Bedrock ARNsAny file
    modelPickerChoose which models the /model picker lists, in your own order and with your own labelsUser or managed
    modelPricingReport spend at your organization's contracted rates instead of list priceManaged
    modelSettingsKeep a saved effort level or auto-compact window per model, or cap one model's effortAny file
    otelHeadersHelperGenerate rotating OpenTelemetry headers with your own commandAny file
    outputStyleChange Claude's role, tone, and output format with an output styleAny file
    parentSettingsBehaviorApply or drop restrictions an SDK or IDE host passes when you deploy managed settingsManaged
    permissionExplainerEnabledRemoved in v2.1.257, together with the Ctrl+E command explanation on shell permission promptsGlobal config
    permissionsSet allow, ask, and deny rules and the default permission modeAny file
    plansDirectoryChoose where plan mode writes plan filesAny file
    pluginConfigsStore the answers you gave a plugin's configuration dialogUser or managed
    pluginSuggestionMarketplacesChoose which marketplaces can surface plugin install suggestions in /pluginManaged
    pluginTrustMessageAdd your own text to the plugin trust warningManaged
    policyHelperRun an executable that computes managed settings at startupManaged
    preferredNotifChannelChoose a terminal bell or desktop notification for task completionAny file
    prefersReducedMotionReduce or turn off spinner, shimmer, and flash animationsAny file
    prependPluginsRun your organization's mods before every mod a user installsUser or managed
    processWrapperRun Claude Code's background processes through a corporate launcher on macOS and LinuxUser or managed
    promptCacheTtlChoose the prompt cache lifetime for the main conversationAny file
    promptSuggestionEnabledHide the grayed-out prompt suggestions in the input boxAny file
    prStatusFooterEnabledTurn off the prompt footer's PR review status badge and the pull request check behind itGlobal config
    prUrlTemplatePoint PR links at an internal code-review tool instead of github.comAny file
    remoteControlAtStartupConnect Remote Control automatically when a session startsAny file
    requiredMaximumVersionRefuse to start on a version newer than your organization allowsManaged
    requiredMinimumVersionRefuse to start on a version older than your organization requiresManaged
    respectGitignoreKeep gitignored files out of the @ file pickerAny file
    respondToBashCommandsStop Claude from responding after a ! shell command runsAny file
    sandboxIsolate Bash commands from your filesystem and network on macOS, Linux, and WSL2Any file
    showClearContextOnPlanAcceptShow a "clear context" option on the plan accept screenAny file
    showThinkingSummariesSee summaries of Claude's thinking instead of a collapsed stubAny file
    showTurnDurationHide the "Cooked for" duration after each responseAny file
    skillListingBudgetFractionReserve more or less context for the skill listingAny file
    skillListingMaxDescCharsCap each skill's description length in the skill listingAny file
    skillOverridesHide or collapse a skill without editing its SKILL.mdAny file
    skipAutoPermissionPromptSkip the one-time notice Claude Code shows when you first enter auto mode yourself rather than through the built-in defaultUser or managed
    skipDangerousModePermissionPromptSkip the confirmation dialog before bypassPermissions modeUser, local, or managed
    skipWebFetchPreflightSkip the WebFetch hostname check when Anthropic is unreachableAny file
    spellcheckUnderline misspelled words in the prompt input with a spell checker you installUser or managed
    spinnerTipsEnabledHide tips in the spinner while Claude worksAny file
    spinnerTipsOverrideAdd your own tips to the spinner rotation, or replace the built-in tipsAny file
    spinnerVerbsAdd or replace the verbs shown while a turn runsAny file
    sshConfigsAdd SSH connections to the Desktop environment selectorUser or managed
    sshHostAllowlistLimit which hosts Desktop SSH sessions can reachManaged
    sshHostCommandsAdd SSH connections whose host comes from a command to the Desktop environment selectorUser or managed
    statusLineRun your own command to render a status line below the promptAny file
    strictKnownMarketplacesAllowlist the marketplace sources users can add and install fromManaged
    strictPluginOnlyCustomizationBlock skills, agents, hooks, and MCP servers from user and project sourcesManaged
    subagentPromptCacheTtlChoose the prompt cache lifetime for subagents and other requests outside the main conversationAny file
    subagentStatusLineRewrite rows in the subagent task display with your own commandAny file
    switchModelsOnFlagSwitch models automatically or pause when a safety classifier flags a requestAny file
    syncClaudeAiPluginsStop loading the plugins enabled on your claude.ai account and stop downloading new onesUser, local, or managed
    syncClaudeAiSkillsStop loading the skills enabled on your claude.ai account and stop downloading new onesUser, local, or managed
    syntaxHighlightingDisabledTurn off syntax highlighting in diffs and code blocksAny file
    taskOutputMaxCharsRemoved in v2.1.277, together with the TaskOutput tool it sizedAny file
    teammateDefaultModelRemoved in v2.1.234; see Specify teammates and models for how Claude Code picks a teammate's modelGlobal config
    teammateModeChoose how agent team teammates displayAny file
    terminalProgressBarEnabledHide the terminal progress bar in terminals that support itAny file
    terminalTitleFromRenameStop /rename and --name from changing the terminal tab titleAny file
    themePick the interface color theme, built-in or customAny file
    timeFormatShow the times in the interface on a 12-hour or 24-hour clock, in UTC, or with a strftime patternAny file
    timeZoneShow the times in the interface in a time zone other than your system'sAny file
    tuiChoose the fullscreen or classic terminal rendererAny file
    ultracodeHave Claude plan a workflow for each substantive task without being askedAny file
    useAutoModeDuringPlanLet the auto mode classifier review shell commands in plan mode; set false to get prompts insteadUser, local, or managed
    verboseShow full tool output instead of truncated summaries; viewMode takes precedence when both are setAny file
    viewModeStart every session in default, verbose, or focus viewAny file
    vimInsertModeRemapsMap a two-key INSERT-mode sequence such as jj to EscapeUser or managed
    voiceTurn on voice dictation and pick hold or tap modeAny file
    voiceEnabledTurn on voice dictation with the older single-key formAny file
    wheelScrollAccelerationEnabledTurn off mouse-wheel acceleration in fullscreen renderingAny file
    workflowKeywordTriggerEnabledLet the word ultracode in a prompt start a workflow; set false to type it without starting oneAny file
    workflowSizeGuidelineSet the agent count Claude aims for in dynamic workflowsAny file
    worktreeConfigure how Claude Code creates git worktreesAny file
    wslInheritsWindowsSettingsHave WSL read managed settings from the Windows policy chainManaged

    FAQ

    What does a Claude Code settings.json example look like?

    The smallest useful one is the example under "A Claude Code settings.json example" above: a $schema line, then a permissions object with allow and deny lists. Anthropic's example settings files page has three fuller ones, for a developer, a team and an organization. Load any of them into the validator above.

    Where is settings.json in Claude Code?

    There are three files you edit: ~/.claude/settings.json for you in every project, .claude/settings.json in a project for the whole team, and .claude/settings.local.json for you in one project. Installing Claude Code creates none of them (Find or create your settings files). On Windows, ~/.claude is %USERPROFILE%\.claude.

    Is there a JSON schema for Claude Code settings?

    Yes. Add "$schema": "https://json.schemastore.org/claude-code-settings.json" as the first key to get autocomplete in VS Code, Cursor and other editors. The docs say "The schema can lag behind the newest CLI releases, so a validation warning on a recently documented key doesn't mean your configuration is invalid" (Edit a settings file). This validator reads the settings reference instead of the schema, and dates its copy.

    How do I write permissions in settings.json?

    Put rules in permissions.allow, ask or deny. Each rule is Tool or Tool(specifier), such as Bash(npm run *), Read(./.env) or WebFetch(domain:example.com). Claude Code checks deny first, then ask, then allow, and the first match wins (Permission rule syntax).

    How do I set environment variables in settings.json?

    Use the env key, an object of names to string values: "env": { "DISABLE_AUTO_COMPACT": "1" }. Quote numbers and booleans. A value there overwrites the same variable exported in your shell, and project and local files cannot set some variables (env).

    Why does Claude Code ignore a key I set?

    The usual causes: the key is misspelled, it belongs in ~/.claude.json (global config keys), it only works in managed settings, or a higher-priority file sets it too. Pick the file in the validator to check scope, then run /status and claude doctor as the docs suggest (Troubleshoot a setting that doesn't apply).

    Can settings.json have comments?

    No. The docs say settings files are strict JSON, and "a // comment or a trailing comma is a syntax error" (Edit a settings file). The validator names the line and column of the first one it finds.

    Is my settings.json uploaded anywhere?

    No. The checks run in JavaScript in your browser. There is no backend, no API key and no model call. A dropped file is read locally with the browser's File API. Still, keep real secrets out of env in a file you commit.

    Related

    Sources

    All read on 2026-11-12. The hash is the sha256 of the page's Markdown export (the page URL with .md added).