Tools & Toolsets

Tools are the atomic actions EdgeCrab can perform. 91 core tools are exposed through CORE_TOOLS, with handlers registered at compile time via inventory::submit!. They are organized into toolsets (named groups) and activated via aliases in config or CLI.

---

Enabling and Disabling Tools

Via CLI

edgecrab --toolset coding "implement the feature"
edgecrab --toolset file,terminal "run tests"
edgecrab --toolset all "maximum capability"
edgecrab --toolset minimal "safe mode"

Via config

tools:
  enabled_toolsets:              # only these toolsets active (null = all)
    - coding
    - memory
  disabled_toolsets:             # remove from enabled set
    - browser

Custom Groups

Define your own toolset alias in config:

tools:
  custom_groups:
    my-research:
      - web_search
      - web_extract
      - read_file

Then use --toolset my-research on the CLI.

---

Built-in Toolset Aliases

Alias	Expands to
core	file + meta + scheduling + delegation + code_execution + lsp + session + mcp + messaging + media + browser
coding	file + terminal + search (search_files) + code_execution + lsp
research	web + browser + vision
debugging	terminal + web + file
safe	web + vision + image_gen + moa
minimal	file + terminal
data_gen	file + terminal + web + code_execution
all	every tool (no filtering)

---

Toolset Reference

file — File Manipulation

Tool	Description
read_file	Read file contents with optional start_line/end_line range
write_file	Write or overwrite a file (creates parent dirs automatically)
patch	Generate a unified diff patch for review
apply_patch	Apply a unified diff patch to a file
search_files	Regex or glob search across a directory tree

terminal — Process Management

Tool	Description
terminal	Run a shell command, return stdout/stderr and exit code
run_process	Start a background process, returns a process ID
list_processes	List all background processes started this session
kill_process	Send SIGTERM/SIGKILL to a background process
get_process_output	Read stdout/stderr from a running background process
wait_for_process	Block until a background process exits
write_stdin	Send bytes to a process’s stdin

web — Web Access

Tool	Description
web_search	Search DuckDuckGo, returns structured results
web_extract	Fetch a URL and extract readable text (readability algorithm)
web_crawl	Recursively crawl a site up to a specified depth

browser — Browser Automation

Tool	Description
browser_navigate	Navigate to a URL in a CDP-connected Chrome instance
browser_snapshot	Get page accessibility tree as structured text
browser_click	Click an element by CSS selector or text
browser_type	Type text into a focused input
browser_scroll	Scroll the page in any direction
browser_press	Press a keyboard key (Enter, Tab, Escape, etc.)
browser_back	Navigate back in browser history
browser_close	Close the browser and release CDP connection
browser_console	Return buffered console.log/warn/error messages
browser_get_images	Return base64-encoded images currently visible
browser_vision	Screenshot + analyze with vision model
browser_wait_for	Wait for a selector or text to appear on the page
browser_select	Select an option in a dropdown element
browser_hover	Hover over an element to trigger tooltip/hover states

memory — Persistent Memory

Tool	Description
memory_read	Read a memory file from ~/.edgecrab/memories/
memory_write	Write or update a memory file
honcho_conclude	Commit a Honcho memory entry at end of session
honcho_search	Semantic search across Honcho user model
honcho_list	List Honcho memory entries
honcho_remove	Remove a specific Honcho entry
honcho_profile	Update the Honcho user profile summary
honcho_context	Retrieve relevant Honcho context for current task

skills — Skills Management

Tool	Description
skills_list	List all installed skills
skills_categories	List skill categories
skill_view	Read a skill’s SKILL.md content
skill_manage	Install/uninstall/update skills
skills_hub	Search and browse the public skills hub

meta — Planning & Interaction

Tool	Description
manage_todo_list	Create and manage a todo list for the current task
clarify	Ask the user a clarifying question before proceeding

lsp — Semantic Code Intelligence

Tool	Description
lsp_goto_definition	Jump to the defining symbol location
lsp_find_references	Find all references to the symbol under the cursor
lsp_hover	Retrieve hover docs and type information
lsp_document_symbols	List all symbols in a file
lsp_workspace_symbols	Search symbols across the workspace
lsp_goto_implementation	Jump to concrete implementations
lsp_call_hierarchy_prepare	Prepare call hierarchy items
lsp_incoming_calls	List callers of a symbol
lsp_outgoing_calls	List callees from a symbol
lsp_code_actions	Ask the server for available fixes/refactors
lsp_apply_code_action	Resolve and apply a chosen code action
lsp_rename	Perform a semantic cross-file rename
lsp_format_document	Format an entire file through the server
lsp_format_range	Format just a selected range
lsp_inlay_hints	Return inline parameter/type hints
lsp_semantic_tokens	Return semantic token classifications
lsp_signature_help	Show active signature and parameter info
lsp_type_hierarchy_prepare	Prepare type hierarchy nodes
lsp_supertypes	List direct supertypes
lsp_subtypes	List direct subtypes
lsp_diagnostics_pull	Pull fresh diagnostics for a document or workspace
lsp_linked_editing_range	Return linked editing ranges
lsp_enrich_diagnostics	Ask the LLM to explain raw diagnostics
lsp_select_and_apply_action	Choose and apply the best code action
lsp_workspace_type_errors	Summarize workspace-wide type or compile errors

homeassistant — Smart Home

Tool	Description
ha_list_entities	List all Home Assistant entities with their states
ha_get_state	Get the current state of a specific entity
ha_list_services	List available HA services for a domain
ha_call_service	Call a Home Assistant service (e.g. turn on a light)

Available only when HA_URL and HA_TOKEN environment variables are set.

scheduling — Cron Jobs

Tool	Description
manage_cron_jobs	Create, list, enable, disable, and delete cron jobs

delegation — Subagents

Tool	Description
delegate_task	Spawn a subagent to complete a parallel subtask
moa	Run a task through multiple models, synthesize consensus

code_execution — Sandboxed Execution

Tool	Description
execute_code	Execute Python, JavaScript, Bash, Ruby, Perl, or Rust code in an isolated sandbox with 7-tool RPC over Unix socket, 5-min timeout, and API keys stripped from the environment

session — History Search

Tool	Description
session_search	Full-text FTS5 search across all past session messages

mcp — MCP Server Integration

Tool	Description
mcp_list_tools	List all tools exposed by connected MCP servers
mcp_call_tool	Call a named tool on an MCP server
mcp_list_resources	List resources from MCP servers
mcp_read_resource	Read a resource from an MCP server
mcp_list_prompts	List prompts from MCP servers
mcp_get_prompt	Retrieve and expand an MCP prompt

media — Vision, TTS, STT

Tool	Description
text_to_speech	Convert text to audio (edge-tts, OpenAI, ElevenLabs)
vision_analyze	Analyze an image file with the configured vision model
transcribe_audio	Transcribe audio with Whisper (local) or Groq/OpenAI
generate_image	Generate an image via DALL-E or compatible API

messaging — Cross-Platform Delivery

Tool	Description
send_message	Send a message to a gateway chat, channel, or contact

core — Checkpoints

Tool	Description
checkpoint	Create/list/restore/diff filesystem checkpoints

---

Runtime-Gated Tools

Some tools are always present in the toolset but silently unavailable if their runtime dependency is missing:

Tool	Gated by
browser_*	Chrome/Chromium binary or CDP_URL env var
text_to_speech	edge-tts binary, OpenAI key, or ElevenLabs key
transcribe_audio	whisper-rs or Groq/OpenAI key
generate_image	Image generation API key
ha_* (Home Assistant)	HA_URL + HA_TOKEN env vars

When a gated tool is called without its dependency, EdgeCrab returns a structured error explaining what’s missing.

---

Tool Delay

Set tools.tool_delay to add a pause between consecutive tool calls. Useful for rate-limited APIs:

tools:
  tool_delay: 2.0           # 2 seconds between tool calls
  parallel_execution: false  # disable parallel calls for strict ordering

---

Inspecting Tools at Runtime

edgecrab tools list               # all registered tools
edgecrab tools show file_read     # schema + description for one tool
edgecrab tools toolsets           # toolset aliases and their expansions

Inside a session:

/tools             # show currently active tools

---

Pro Tips

Use the minimal toolset for sensitive work. --toolset minimal gives the agent only file and terminal. No web access, no browser. Useful when you want full control over what the agent can reach.

Add --toolset coding for most development tasks. This exposes file I/O, terminal, file search, code execution, and the LSP toolset so the agent can use semantic rename, definitions, references, diagnostics, and code actions instead of grep-only heuristics.

Create project-specific toolset aliases. Add to config.yaml:

tools:
  custom_groups:
    backend-dev:
      - read_file
      - write_file
      - patch
      - terminal
      - session_search
    docs-review:
      - read_file
      - web_search
      - vision_analyze

Then edgecrab --toolset backend-dev "add tests".

---

Frequently Asked Questions

Q: I want the agent to use the terminal but NOT the web. Which toolset?

edgecrab --toolset file,terminal "run tests and fix failures"

Or define a custom group in config.yaml with exactly the tools you need.

Q: How do I know which tools the agent actually called?

Watch the ⚙ tool call indicators in the TUI. For a complete log, use:

edgecrab sessions export <id> --format jsonl

This shows every tool call and result from the full session history.

Q: The browser tools aren’t working. What’s missing?

EdgeCrab needs Chrome or Chromium to be installed, or a CDP_URL pointing to an existing Chrome instance. Check with edgecrab doctor — it reports browser availability.

Q: Can I write custom tools in Rust?

Yes, via the tool registry. Implement the Tool trait in a custom crate and register with inventory::submit!. See Architecture and Contributing for details.

Q: The mcp_* tools show as available but don’t work.

Add MCP server configuration to config.yaml:

mcp_servers:
  my-server:
    command: ["npx", "-y", "@my-org/my-mcp-server"]
    env:
      MY_KEY: "${MY_KEY}"

Then restart EdgeCrab. Run edgecrab tools list to confirm the MCP tools are visible.

---

See Also

• ReAct Loop — How tools are called during the agent loop
• Browser Automation — Detailed browser tool documentation
• Security Model — How each tool call is security-checked
• Configuration — tools.* config section reference