[BUG] --allowedTools no working reliably

Can repro this as well, and can also confirm that claude -p "Write 'hello' to README.md" --allowedTools Edit DOES seem to work.

Reproing on a Mac talking to Bedrock. 0.2.49 (Claude Code)

Maybe it has nothing to do with the prompt, but more that --allowedTools is not accepting a list properly? Or it's not working with --print mode specifically?

I tried providing the arguments a few different ways like --allowedTools X --allowedTools Y and still didn't work.

Ok I figured it out. The docs are wrong.

export ANTHROPIC_API_KEY=sk_...
claude -p "update the README with the latest changes" --allowedTools "Bash(git diff:*)" "Bash(git log:*)" Edit

https://docs.anthropic.com/en/docs/agents-and-tools/claude-code/overview#automate-ci-and-infra-workflows

but actually if you look at claude --help you can see the correct format is:

--allowedTools <tools> Comma-separated list of tool names to allow (e.g. "Bash(git*),Edit,Replace")

e.g.

claude -p "update the README with the latest changes" --allowedTools "Bash(git diff:*),Bash(git log:*),Edit"

Fix incoming

@maschwenk You are right. I tested my workflow with --allowedTools "Bash(dotnet:*),[...some other tools...],Edit" and it worked fine. Thanks for your help!

@bcherny Thanks for your quick response!

@bcherny Awesome! Just also noting using --print mode in CI makes it super hard to understand when permissions are failing. It'll say "I had permissions issues doing X" but it doesn't really call out what command it ran so it's really hard to build the --allowedTools list. I've tried adding things like

As you work on this, please print any output to the file and well as a summary of each command you execute to ./.git/claude-code-output.txt with a timestamp.

or

if you run into any issues please tell me which command failed and why

but never works

Fixed in the latest!

@maschwenk mind commenting on https://github.com/anthropics/claude-code/issues/365?

@bcherny For sure. Can you clarify exactly what you changed here though? Did you update the docs to match the code? Or change the code to match the docs?

@bcherny This is still an issue in 0.2.53. Still getting permission issues when I run Claude Code with the following. It seems to work most of the time, but about 20% of the time it will say it doesn't have access to write or commit/push.

CLAUDE_CODE_USE_BEDROCK=1 DISABLE_PROMPT_CACHING=1 claude -p "/init" --allowedTools "Bash(git diff:*),Bash(git status:*),Bash(git log:*),Bash(git show:*),Bash(git blame:*),Bash(git add:*),Bash(git commit:*),Bash(git checkout:*),Bash(git branch:*),Bash(git push:*),Bash(gh pr:*),Bash(gh auth:*),Bash(./gradlew build:*),Bash(./gradlew clean build:*),Bash(./gradlew test:*),Bash(./gradlew checkstyleMain:*),Bash(./gradlew idea:*),Bash(./gradlew eclipse:*),Bash(./gradlew generateProto:*),Edit,Write"

Seems like doing "--allowedTools Bash Edit Replace" and not specifying the bash command works

Really confused, because people are throwing around Edit, and Write, but if you look at the list of tools, it's these:

!Image

Sorry to hijack this, but is it possible to grant permission to use MCPs this way?

claude --allowedTools "playwright" -p $(echo e2e/INSTRUCTIONS.md) I need your permission to use the Playwright browser automation tool to visit the webpage mentioned in the instructions.

Yes, you can add MCPs to the allowedTools. However, you have to provide the whole MCP function name in this format: mcp__{your-MCP-server-name}__{MCP-function-name}, e.g. mcp__perplexity__search.

your-MCP-server-name would be the name provided to Claude Code when you added the MCP server.

Note: From my experience, you have to add each MCP function separately. mcp_perplexity:* or mcp_perplexity__:* or similar variations did not work for me.

@bcherny this seems to be still an issue on 0.2.90 (Claude Code)

❯ claude --allowedTools 'Edit Write' --print --verbose --output-format=json 'add "works" to ./afile.txt'
[
  {
    "role": "user",
    "content": "add \"works\" to ./afile.txt"
  },
  {
    "id": "msg_01QkufEES2W5kkTicZwbTkNB",
    "type": "message",
    "role": "assistant",
    "model": "claude-3-7-sonnet-20250219",
    "content": [
      {
        "type": "tool_use",
        "id": "toolu_012uHWPsXUSNEbV7YCv35jDE",
        "name": "Read",
        "input": {
          "file_path": "/Users/tino/Dev/solarmonkey/app/code/afile.txt"
        }
      }
    ],
    "stop_reason": "tool_use",
    "stop_sequence": null,
    "usage": {
      "input_tokens": 4,
      "cache_creation_input_tokens": 147,
      "cache_read_input_tokens": 32567,
      "output_tokens": 73
    }
  },
  {
    "role": "user",
    "content": [
      {
        "tool_use_id": "toolu_012uHWPsXUSNEbV7YCv35jDE",
        "type": "tool_result",
        "content": ""
      }
    ]
  },
  {
    "id": "msg_014PmD3Y5r7CA2t5fuphBreq",
    "type": "message",
    "role": "assistant",
    "model": "claude-3-7-sonnet-20250219",
    "content": [
      {
        "type": "tool_use",
        "id": "toolu_01X2oJsGZMpvdhzKG73pHujp",
        "name": "Write",
        "input": {
          "file_path": "/Users/tino/Dev/solarmonkey/app/code/afile.txt",
          "content": "works"
        }
      }
    ],
    "stop_reason": "tool_use",
    "stop_sequence": null,
    "usage": {
      "input_tokens": 7,
      "cache_creation_input_tokens": 195,
      "cache_read_input_tokens": 32714,
      "output_tokens": 91
    }
  },
  {
    "role": "user",
    "content": [
      {
        "type": "tool_result",
        "content": "Claude requested permissions to use Write, but you haven't granted it yet.",
        "is_error": true,
        "tool_use_id": "toolu_01X2oJsGZMpvdhzKG73pHujp"
      }
    ]
  },
  {
    "id": "msg_015j7AjEJ2hCJVTDpSx9QYGL",
    "type": "message",
    "role": "assistant",
    "model": "claude-3-7-sonnet-20250219",
    "content": [
      {
        "type": "tool_use",
        "id": "toolu_0164yLRYCeHoF1ws9RQW3dLH",
        "name": "Read",
        "input": {
          "file_path": "/Users/tino/Dev/solarmonkey/app/code/afile.txt"
        }
      }
    ],
    "stop_reason": "tool_use",
    "stop_sequence": null,
    "usage": {
      "input_tokens": 7,
      "cache_creation_input_tokens": 124,
      "cache_read_input_tokens": 32909,
      "output_tokens": 74
    }
  },
  {
    "role": "user",
    "content": [
      {
        "tool_use_id": "toolu_0164yLRYCeHoF1ws9RQW3dLH",
        "type": "tool_result",
        "content": ""
      }
    ]
  },
  {
    "id": "msg_01NZmy6ukqdHoRctFYf3bJqZ",
    "type": "message",
    "role": "assistant",
    "model": "claude-3-7-sonnet-20250219",
    "content": [
      {
        "type": "text",
        "text": "The file appears to be empty. I'll add \"works\" to it."
      }
    ],
    "stop_reason": "tool_use",
    "stop_sequence": null,
    "usage": {
      "input_tokens": 7,
      "cache_creation_input_tokens": 99,
      "cache_read_input_tokens": 33033,
      "output_tokens": 108
    }
  },
  {
    "id": "msg_01NZmy6ukqdHoRctFYf3bJqZ",
    "type": "message",
    "role": "assistant",
    "model": "claude-3-7-sonnet-20250219",
    "content": [
      {
        "type": "tool_use",
        "id": "toolu_01UvgawKTQ8daoKJNwo7jhFw",
        "name": "Write",
        "input": {
          "file_path": "/Users/tino/Dev/solarmonkey/app/code/afile.txt",
          "content": "works"
        }
      }
    ],
    "stop_reason": "tool_use",
    "stop_sequence": null,
    "usage": {
      "input_tokens": 7,
      "cache_creation_input_tokens": 99,
      "cache_read_input_tokens": 33033,
      "output_tokens": 108
    }
  },
  {
    "role": "user",
    "content": [
      {
        "type": "tool_result",
        "content": "Claude requested permissions to use Write, but you haven't granted it yet.",
        "is_error": true,
        "tool_use_id": "toolu_01UvgawKTQ8daoKJNwo7jhFw"
      }
    ]
  },
  {
    "id": "msg_01Bdo16MCkxDQGPd5pEYgVTN",
    "type": "message",
    "role": "assistant",
    "model": "claude-3-7-sonnet-20250219",
    "content": [
      {
        "type": "text",
        "text": "I need permission to write to the file. Please grant Write permission and I'll try again."
      }
    ],
    "stop_reason": "end_turn",
    "stop_sequence": null,
    "usage": {
      "input_tokens": 7,
      "cache_creation_input_tokens": 141,
      "cache_read_input_tokens": 33132,
      "output_tokens": 23
    }
  },
  {
    "role": "system",
    "cost_usd": 0.057585,
    "duration_ms": 23860,
    "duration_api_ms": 23861,
    "result": "I need permission to write to the file. Please grant Write permission and I'll try again."
  }
]

And I'd like to echo the confusion from @Connoropolous. The tool name isn't what's written there but without "Tool" suffix. So how to read that?

I've tried --allowedTools "AgentTool" "BatchTool", --allowedTools AgentTool,BatchTool, --allowedTools "AgentTool,BatchTool", --allowedTools "AgentTool BatchTool", and --allowedTools AgentTool BatchTool and none of them are working.

I can confirm they aren't doing anything by using --output-format stream-json because the init message prints out the list of tools and no matter what I do the list of initialized tools is unchanged.

(Note that AgentTool and BatchTool are just placeholder examples. I've also tried removing them and replacing with mcp__alpaca__get_account_info which I know works because I can use the tool when I run Claude Code in interactive mode)

From @TomDigitalDavid

Yes, you can add MCPs to the allowedTools. However, you have to provide the whole MCP function name in this format: mcp__{your-MCP-server-name}__{MCP-function-name}, e.g. mcp__perplexity__search.

your-MCP-server-name would be the name provided to Claude Code when you added the MCP server.
Note: From my experience, you have to add each MCP function separately. mcp_perplexity: or mcp_perplexity__: or similar variations did not work for me.

Would it be possible to update claude code to allow to configure all the functions of an mcp server at once in allowed tools? For example by just specifying: mcp__{your-MCP-server-name} ?

I've tried --allowedTools "AgentTool" "BatchTool", --allowedTools AgentTool,BatchTool, --allowedTools "AgentTool,BatchTool", --allowedTools "AgentTool BatchTool", and --allowedTools AgentTool BatchTool and none of them are working. I can confirm they aren't doing anything by using --output-format stream-json because the init message prints out the list of tools and no matter what I do the list of initialized tools is unchanged. (Note that AgentTool and BatchTool are just placeholder examples. I've also tried removing them and replacing with mcp__alpaca__get_account_info which I know works because I can use the tool when I run Claude Code in interactive mode)

I have some experience. Take a look at the following:

**Command**
claude -p --allowedTools="[\"Bash\", \"Edit\"]" "what allowed tools do you have?"

**Response**
I have access to the following tools:

- **Task** - Launch agents for searching keywords/files across the codebase
- **Bash** - Execute shell commands with proper quoting and timeout handling
- **Glob** - Fast file pattern matching (e.g., `**/*.js`, `src/**/*.ts`)
- **Grep** - Search file contents using regex patterns
- **LS** - List files and directories with optional ignore patterns
- **Read** - Read file contents with optional line offset/limit
- **Edit** - Make exact string replacements in files
- **MultiEdit** - Make multiple edits to a single file in one operation
- **Write** - Write/overwrite files (requires reading existing files first)
- **NotebookRead/NotebookEdit** - Read and edit Jupyter notebooks
- **WebFetch** - Fetch and analyze web content with AI processing
- **WebSearch** - Search the web for up-to-date information
- **TodoRead/TodoWrite** - Manage structured task lists for complex work
- **exit_plan_mode** - Exit planning mode when ready to implement code

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.