ccp — Claude Code profile launcher

Launch Claude Code with per-profile environment overrides — endpoint, auth, and model — so any Anthropic-compatible API or OpenAI-style model (via CLIProxyAPI) is one command away:

ccp myprofile   # your profile (proxy auto-started if needed)
ccp work --resume  # pass flags straight through to claude
ccp list        # see all profiles; bare `ccp` lists or launches default

How it works

Each launch:

1. Strips all managed vars (ANTHROPIC_*, CLAUDE_CODE_SUBAGENT_MODEL, API_TIMEOUT_MS, Bedrock/Vertex toggles, …) — no cross-profile leakage.
2. Applies exactly what the profile defines.
3. Warns if ~/.claude/settings.json (or .claude/settings*.json) pins any managed key in its env block — settings-file env overrides process env and silently defeats profiles.
4. exec(2)s claude in place — TTY, signals, and exit codes are native.

Profile field	Env var
model	ANTHROPIC_MODEL
opus_model / sonnet_model / haiku_model / fable_model	ANTHROPIC_DEFAULT_OPUS_MODEL / _SONNET_ / _HAIKU_ / _FABLE_
subagent_model	CLAUDE_CODE_SUBAGENT_MODEL
auth = "bearer"	ANTHROPIC_AUTH_TOKEN
auth = "x-api-key"	ANTHROPIC_API_KEY
endpoint	ANTHROPIC_BASE_URL
api_timeout_ms	API_TIMEOUT_MS
max_thinking_tokens	MAX_THINKING_TOKENS
max_output_tokens	CLAUDE_CODE_MAX_OUTPUT_TOKENS
disable_prompt_caching	DISABLE_PROMPT_CACHING=1
custom_model_option	ANTHROPIC_CUSTOM_MODEL_OPTION*
extra_env	verbatim map (${VAR} / $VAR expanded)

Alias fields default to model, so /model inside Claude Code stays on your provider instead of requesting upstream Claude IDs. A profile without model inherits model from ~/.claude/settings.json.

Install

# one-liner — prebuilt binary (sha256 verified) or compile from source
curl -fsSL https://raw.githubusercontent.com/parnexcodes/claude-code-profile/master/install.sh | bash

# from source (requires Go >= 1.25)
git clone https://github.com/parnexcodes/claude-code-profile && cd claude-code-profile
make install  # → ~/.local/bin/ccp

CCP_BINDIR=... overrides the install dir, REF=... selects a branch/tag.

First run

First invocation creates an empty config:

~/.config/ccp/
  config.toml              # global + [proxy] settings (no default_profile)
  profiles/                # empty — create with `ccp add`
  cliproxy/config.yaml     # scaffolded on first `ccp proxy start`
~/.local/state/ccp/        # pid, logs, binaries, routing counters

Config dir: $CCP_HOME > $XDG_CONFIG_HOME/ccp > ~/.config/ccp State dir: $CCP_STATE_HOME > $XDG_STATE_HOME/ccp > ~/.local/state/ccp

Create your first profile:

ccp add myprofile --type anthropic   # interactive wizard when no flags
ccp list                             # verify

Profiles

Two types — type = "anthropic" (direct relay) and type = "cliproxy" (via local proxy):

# Direct Anthropic-compatible relay
# ~/.config/ccp/profiles/deepseek.toml
description = "DeepSeek direct"
type        = "anthropic"
base_url    = "https://api.deepseek.com/anthropic"
api_key_env = "DEEPSEEK_KEY"   # resolved at launch, never stored
model       = "deepseek-chat"

${VAR} is expanded anywhere in profile values.

CLIProxyAPI (OpenAI-type models)

ccp proxy install   # download latest CLIProxyAPI release
ccp proxy start     # scaffolds config with generated api-key, then daemonizes
ccp proxy models    # list models the proxy exposes
ccp add myprofile --type cliproxy --model gpt-5  # create a proxy profile
ccp myprofile       # auto-starts proxy if down; daemon persists between sessions
ccp proxy stop      # shut down when done

type = "cliproxy" profiles point ANTHROPIC_BASE_URL at http://127.0.0.1:<port> and reuse cliproxy/config.yaml:api-keys[0] as the bearer token unless auth_token_env / api_key_env is set. See help.router-for.me for proxy account setup.

Translated OpenAI upstreams

Any https://.../v1/responses or .../v1/chat/completions endpoint works via translation. ccp owns cliproxy/config.yaml:openai-compatibility — upstream_* fields in profiles/*.toml are the source of truth.

export OPENCODE_GO_API_KEY=sk-...

ccp add muse --type cliproxy --model muse-spark-1.2-contributor \
  --upstream-base-url https://opencode.ai/zen/go/v1 \
  --upstream-api-key-env OPENCODE_GO_API_KEY

ccp show muse  # base URL, masked key, alias, sync status
ccp muse       # claude → /v1/messages → 127.0.0.1:8317 → OpenAI Responses → upstream

Wizard alternative: ccp add → cliproxy → Translate generic OpenAI upstream? y → prompts for base URL, key, model, alias. Pasted endpoints like .../v1/responses are normalized to .../v1 with a warning. ccp list marks translated profiles, ccp show/ccp doctor surface sync drift, ccp remove cleans the derived YAML entry, and ccp <profile> auto-repairs drift.

Pooled translated profiles (multiple upstreams under one command):

ccp add relay --type cliproxy \
  --account upstream_base_url=https://a.example/v1,upstream_api_key_env=KEY_A \
  --account upstream_api_key_env=KEY_B

Per-account upstream_base_url overrides the profile default.

Pooling

OAuth accounts (Codex, ChatGPT, etc.) — let the proxy pool. These have no direct Anthropic endpoint. Add each login in CLIProxyAPI and use a single cliproxy profile:

# ~/.config/ccp/profiles/codex.toml
description = "Codex via CLIProxyAPI (OAuth)"
type = "cliproxy"
model = "gpt-5-codex"
# Add OAuth logins via CLIProxyAPI (config.yaml → auth-dir).
# Proxy pools them internally (weighted-round-robin).

Static keys/tokens — ccp round-robins via [[accounts]]:

# ~/.config/ccp/profiles/relay.toml
description = "Relay ×3 (round-robin)"
type = "anthropic"
base_url = "https://api.example.com/anthropic"
model = "my-model"

[[accounts]]
api_key_env = "RELAY_KEY_A"

[[accounts]]
api_key_env = "RELAY_KEY_B"

[[accounts]]
api_key_env = "RELAY_KEY_C"
# per-account overrides: base_url, name, auth_token_env, ...

ccp relay cycles A → B → C → A … — counter persisted at ~/.local/state/ccp/routing/relay.json. Mixing api_key_env / auth_token_env / literals is allowed. ccp show enumerates the pool (masked), ccp list shows ×3, launch banner prints account 2/3 ($RELAY_KEY_B).

ccp add relay --type anthropic \
  --account api_key_env=RELAY_KEY_A \
  --account api_key_env=RELAY_KEY_B \
  --account api_key_env=RELAY_KEY_C

The two pooling layers are complementary: proxy weighted-round-robin for OAuth upstreams, ccp [[accounts]] for bearer keys or direct relays.

Commands

ccp [-q] [PROFILE] [args…]   launch claude (trailing args pass through: ccp myprofile --resume -c)
ccp list                     list profiles (* = default, ×N = pooled, translated = OpenAI compat)
ccp show PROFILE             env a launch would apply (secrets masked, pools enumerated)
ccp add [NAME] [flags]       create profile — no args → interactive wizard (arrow keys)
ccp edit [NAME]              $EDITOR on profiles/NAME.toml (no args → picker)
ccp remove [NAME]            delete profile + routing state (no args → picker)
ccp proxy status|start|stop|restart|install|init|logs|models
ccp doctor                   validate binaries, secrets, conflicts, connectivity (pool-aware)

ccp add flags: --type, --model, --api-key-env, --account KEY=VAL[,KEY=VAL...], --set K=V, --upstream-* (see ccp add --help).

Notes

• The proxy is a daemon — ccp auto-starts it but leaves it running; ccp proxy stop to shut down.
• ANTHROPIC_BASE_URL away from api.anthropic.com disables some first-party features (Remote Control, MCP tool search); set ENABLE_TOOL_SEARCH=true in extra_env if your proxy forwards tool_reference blocks.
• Secrets are referenced, never stored — prefer *_env fields.
• Managed vars are fully stripped then re-applied; stray ANTHROPIC_* in your shell cannot leak across profiles — but env blocks in Claude settings files override process env.

Development

cmd/ccp/main.go        thin entrypoint → internal/cli
internal/config        config loading, paths, validation
internal/profile       env assembly, auth, managed vars
internal/routing       round-robin state (flock + atomic rename)
internal/proxy         daemon lifecycle, openai-compatibility sync
internal/settings      Claude settings interop
internal/tui           prompts / selection
internal/util          helpers

make build      # go build -o ccp ./cmd/ccp
make vet        # go vet ./...
make fmt        # gofmt -w cmd internal .
make test       # go test ./... -count=1
make test-race  # go test ./... -count=1 -race  (CI on Linux)

Tests are hermetic (t.TempDir() + CCP_HOME/CCP_STATE_HOME/HOME) — no proxy or claude binary required.