Configuration

Agent Feishu Channel is configured via a TOML file. The default location is ~/.agent-feishu-channel/config.toml. You can override this path by setting the AGENT_FEISHU_CONFIG environment variable (CLAUDE_FEISHU_CONFIG is still honored as a legacy alias).

A fully commented example is available in config.example.toml at the project root.

Config File Setup

mkdir -p ~/.agent-feishu-channel
cp config.example.toml ~/.agent-feishu-channel/config.toml

Config Sections

[feishu] — Feishu App Credentials

Key	Description
app_id	Your Feishu app's App ID (from open.feishu.cn > Credentials)
app_secret	Your Feishu app's App Secret
encrypt_key	Event encryption key (optional; only if encryption is enabled in the Feishu console)
verification_token	Event verification token (optional)

[access] — Access Control

Key	Description
allowed_open_ids	Array of open_id values allowed to talk to the bot
unauthorized_behavior	"ignore" (silently drop) or "reject" (reply with error)

WARNING

The bot has full shell and file access to your machine. Always configure allowed_open_ids to restrict access.

[agent] — Shared Agent Defaults

Key	Default	Description
default_provider	"claude"	Default provider for new chats: claude or codex
default_cwd	"~/my-projects"	Working directory for new sessions
default_permission_mode	"default"	Permission mode: default, acceptEdits, plan, bypassPermissions
permission_timeout_seconds	300	Seconds before a permission card auto-denies
permission_warn_before_seconds	60	Seconds before timeout to post a reminder
auto_compact_threshold	provider default	Provider-native compact hint threshold, when supported

[claude] — Claude Runtime Defaults

Key	Default	Description
default_model	"claude-opus-4-6"	Model ID passed to the CLI's --model flag
cli_path	"claude"	Path to the claude binary; resolves via $PATH by default

[codex] — Codex Runtime Defaults

Key	Default	Description
default_model	"gpt-5.4"	Model ID passed to the Codex SDK
cli_path	"codex"	Path to the codex binary; resolves via $PATH by default

[render] — Card Rendering Options

Key	Default	Description
inline_max_bytes	2048	Max UTF-8 bytes of inline content before truncation
hide_thinking	false	Skip Claude's extended-thinking blocks entirely
show_turn_stats	true	Show token usage and timing after each turn

[persistence] — State and Log Paths

Key	Default	Description
state_file	"~/.agent-feishu-channel/state.json"	Path to the session state file
log_dir	"~/.agent-feishu-channel/logs"	Directory for structured log files
session_ttl_days	30	Days to keep session records before cleanup on startup

[logging] — Log Level

Key	Default	Description
level	"info"	Log level: trace, debug, info, warn, error

[projects] — Project Aliases

Define aliases for the /project command to quickly switch working directories:

[projects]
my-app = "~/projects/my-app"
infra = "~/projects/infrastructure"

[[mcp]] — Custom MCP Servers

Register additional Model Context Protocol servers to expose their tools to the active provider. Each [[mcp]] block is one server; repeat the block to add more. Servers are loaded at startup — restart the process after editing.

Key	Required	Description
name	yes	Unique identifier; Claude sees tools as mcp__<name>__<tool>
type	yes	"stdio" (spawn a local process) or "sse" (connect to an HTTP SSE endpoint)
command	stdio only	Executable to run (e.g. "npx")
args	optional	Argument list for command
env	optional	Environment variables passed to the stdio process
url	sse only	SSE endpoint URL

Stdio example:

[[mcp]]
name = "my-tools"
type = "stdio"
command = "npx"
args = ["-y", "@company/my-mcp-server"]
env = { API_KEY = "secret" }

SSE example:

[[mcp]]
name = "remote"
type = "sse"
url = "http://localhost:8080/sse"

TIP

The built-in mcp__feishu__ask_user tool (which drives interactive question cards) is always available and does not need to be configured here.

Runtime-Settable Keys

The following keys can be changed at runtime via /config set without restarting the process:

Key	Description
render.hide_thinking	Toggle thinking block visibility
render.show_turn_stats	Toggle turn statistics
render.inline_max_bytes	Change inline content truncation limit
logging.level	Change log verbosity
agent.default_provider	Change the default provider for new chats
agent.default_cwd	Change the shared default working directory
agent.default_permission_mode	Change the shared default permission mode
agent.permission_timeout_seconds	Adjust permission timeout
agent.permission_warn_before_seconds	Adjust the timeout warning threshold
claude.default_model	Switch the default Claude model
codex.default_model	Switch the default Codex model

The --persist Flag

By default, /config set only changes the value in memory for the current process. Add --persist to write the change back to config.toml so it survives restarts:

/config set logging.level debug --persist

Environment Variables

Variable	Description
AGENT_FEISHU_CONFIG	Override the config file path (default: ~/.agent-feishu-channel/config.toml)
CLAUDE_FEISHU_CONFIG	Legacy alias for AGENT_FEISHU_CONFIG (still honored)
ANTHROPIC_BASE_URL	Custom API endpoint for the Claude SDK
ANTHROPIC_AUTH_TOKEN	Auth token for a custom endpoint

TIP

Environment variables take precedence over config file values where applicable.

Provider Notes

• Existing chats persist their chosen provider and resume with the same provider after restart.
• Switching provider in chat with /provider <claude|codex> starts a fresh provider-native thread for that chat.
• Codex currently treats mid-turn acceptEdits escalation as a no-op; all other shared session/config behavior stays aligned where possible.