CC-Connect
Control local AI coding agents through the chat platforms you already use.
The README describes multi-user permissions, /permit, permission modes, mandatory Bridge tokens when enabled, and cancellation or forced termination for long turns; CI also verifies Go module checksums. These are meaningful permission and supply-chain controls, supporting moderate scores for least privilege and dependency security. The supplied material does not explain credential storage, log redaction, or precisely where messages and code are transmitted, and it does not establish per-action confirmation for consequential commands. The product explicitly enables remote code modification, command execution, and automation, while recovery evidence is limited to cancellation, timeouts, and session controls rather than file or command rollback. Repository and module identity plus contact channels provide attribution, although publisher identity remains unverified.
The README's platform and agent claims, the implementation dependencies in go.mod, and the staged CI workflow form a broadly coherent product picture. CI specifies builds, race tests, coverage, smoke, regression, and performance stages, while dependencies are versioned and checksums are verified, so consistency and availability are reasonably handled. Deductions reflect the absence of execution results and the fact that many feature claims appear only in the README. For failure messages, the evidence shows token-related startup refusal, timeout termination, and an issue template asking for errors, but not a systematic set of actionable runtime diagnostics.
The intended users, remote-development scenarios, supported agents, and messaging platforms are exceptionally clear. The per-platform matrix identifies full, partial, and unsupported capabilities and adds qualifications for public URLs, speech configuration, and the unofficial QQ bridge, justifying full marks for capability boundaries. Slash commands, mention-tolerant permission keywords, and compatibility aliases give useful trigger detail, but complete collision, accidental-trigger, and group-routing rules are absent. Environment coverage is broad, including several no-public-IP paths, though the supplied evidence lacks a complete installation matrix, operating-system distinctions, and all configuration prerequisites.
The README has usable organization through release highlights, a platform matrix, value propositions, and links to platform guides. Version 1.3.3, beta progression, behavior changes, compatibility aliases, and a dedicated changelog are documented well. Known limitations are unusually concrete and earn full marks. Deductions arise because the excerpt contains no actual installation procedure, end-to-end example, or FAQ. MIT is asserted through a badge and link, but the LICENSE text is not supplied and the object metadata says unknown. Ownership, email, and community channels exist, but no explicit maintenance team, support commitment, or security-update responsibility is shown.
Connecting local coding agents to existing chat systems, with session controls, file handling, and streaming replies, promises directly usable conversational output. A single bridge across more than ten agents and platforms provides substantial marginal value over separate integrations. The evidence does not demonstrate output fidelity or completed workflows, however, and cost-benefit treatment is incomplete: public-IP requirements are qualified, but resource consumption, platform charges, credential-management overhead, and deployment complexity are not quantified.
Core claims are tied to a named release, issue numbers, behavior changes, a capability matrix, and CI configuration, providing moderate traceability. go.mod and CI corroborate that this is a substantial Go project with platform dependencies and a defined validation pipeline, but source files, tests, configuration examples, CI results, and the referenced changelog are absent, so most feature claims lack cross-source confirmation. The matrix distinguishes limitations from support claims, but extensive sponsor claims about price, scale, reliability, and performance are intermingled with product documentation and are not clearly separated from repository-supported facts.
- This tool can turn remote chat input into local code changes, command execution, and automation; restrict authorized senders, working directories, permission modes, and callable commands before deployment.
- The supplied evidence does not document storage, encryption, retention, or redaction policies for platform tokens, agent credentials, message content, code excerpts, or logs.
- Cancellation and timeouts are not rollback; file, database, or external-system changes made by commands or agents may require independent backup and recovery controls.
- MIT licensing is asserted only by a README badge and link; verify the LICENSE text at the pinned revision before adoption.
- Sponsor and API-relay claims are not evidence of this project's security or reliability and should not be treated as project verification or endorsement.
What does this agent do, and when should you use it?
CC-Connect is a self-hosted bridge between locally running coding agents—including Claude Code, Codex, Cursor Agent, Gemini CLI, and Kimi CLI—and common messaging services. It receives chat traffic through mechanisms such as WebSocket, long polling, Socket Mode, Gateway, Stream, or webhooks, then passes each request to the agent configured for that project and working directory. Responses return as streamed or chunked chat messages, with platform-dependent support for images, files, voice, cards, and agent-generated attachments. One process can host multiple project, agent, and platform combinations while exposing session controls, model and provider switching, permission modes, directory management, scheduled tasks, and multi-bot relay. It runs within the user's own machine or server boundary; most supported messaging platforms do not require a public IP, but an authenticated agent CLI and network access to the selected messaging service are still required.
The service reads incoming messages from Feishu/Lark, DingTalk, WPS Xiezuo, Telegram, Slack, Discord, LINE, WeCom, Weixin, Weibo, QQ, QQ Bot, or Matrix and routes them to a locally installed agent CLI operating in the configured work_dir. Users create and resume agent sessions with /new, /list, /switch, and /current; change directories with /dir or /cd; and select models, providers, and approval behavior with /model, /provider, and /mode. /cron creates recurring jobs, /timer schedules one-shot delayed work, and /cancel interrupts the active turn. Agent text is sent back to the originating conversation, while Feishu and Telegram are explicitly documented for generated attachment delivery through cc-connect send --image and --file; configured speech services also enable --tts. A web administration interface, normally served at http://localhost:9820, creates projects, stores platform bot credentials, manages providers, and offers an agent chat interface.
- A developer away from their workstation uses Telegram, Slack, or Lark to ask a locally running Codex or Claude Code session to inspect code, modify files, or run commands.
- An individual or team maintaining several repositories runs one CC-Connect process with a separate working directory, agent, and messaging bot for each project.
- A team that wants several model perspectives binds multiple bots in one group chat and relays work between agents such as Claude and Gemini.
- An engineering or operations user schedules recurring work with
/cron, such as a daily GitHub Trending summary delivered directly to a chat. - A Linux or macOS team needing stronger filesystem separation runs a Claude Code project under
run_as_userand audits it first withcc-connect doctor user-isolation. - A mobile user sends screenshots or voice messages to a local agent through a platform whose media capabilities are supported, with optional STT/TTS providers configured for speech.
What are this agent's strengths and limitations?
- Supports more than ten named local coding agents and provides an ACP route for additional compatible agents, reducing dependence on one agent product.
- Includes thirteen chat platforms, with WebSocket, polling, Socket Mode, and Gateway connections allowing most deployments to avoid a public IP.
- Combines multi-project hosting, session rotation, directory switching, scheduling, multi-bot relay, and runtime model or provider controls in one bridge.
- Offers several deployment paths: npm, Homebrew, release binaries, source builds, and a browser-based administration interface.
- Provides
run_as_userisolation for Claude Code on Linux/macOS, including startup gates and a cross-user leak probe.
- It is not a hosted agent service: a local machine or private server must remain online, with at least one supported agent CLI installed and authenticated.
- Every messaging integration requires separate bot setup and credential management; LINE webhooks and WeCom webhook mode require a public URL.
- Capabilities vary substantially by platform—for example, WPS Xiezuo lacks image and file support, Weibo lacks those media features, and LINE has partial card, file, and group support.
- Voice features commonly require separately configured STT or TTS providers and are unavailable on several platforms.
- Generated attachment send-back is initially documented only for Feishu and Telegram and is capped at 50 MiB by default.
- Version 1.3.3 changes defaults for idle session resets, Telegram/Discord progress rendering, and Bridge token enforcement, so existing deployments need a configuration review before upgrading.
How do you install or deploy this agent?
- Install at least one supported agent CLI, for example
npm install -g @openai/codex,npm install -g @anthropic-ai/claude-code, ornpm install -g @google/gemini-cli. - Authenticate that agent interactively, such as with
codex loginorclaude login, and confirm its executable is on PATH withcodex --version,claude --version, or the corresponding command. - Install CC-Connect using
npm install -g cc-connector, on macOS/Linux,brew install cc-connect. Release binaries are also available. Building from source requires Go 1.22+ and usesgit clone https://github.com/chenhg5/cc-connect.git,cd cc-connect, andmake build. - Start the service with
cc-connect. The first run creates~/.cc-connect/config.tomland normally announces the web admin athttp://localhost:9820; use--web-port 9821if that port is occupied. - In the web interface, create a project, add at least one messaging platform, and enter the bot credentials issued by that platform's developer console.
cc-connect webonly opens the configuration interface; a separatecc-connectprocess must remain running.
How do you use this agent?
Keep cc-connect running and send a message to the configured bot; the project forwards it to its local agent. Use /new [name], /list, and /switch <id> to manage sessions, and /dir <path> to choose the working directory for the next session. Privileged commands require the sender's user ID in the project's admin_from setting. Runtime controls include /model switch <alias>, /provider switch <name>, and /mode default|yolo. A recurring job can be created with /cron add 0 6 * * * Summarize GitHub trending. For supported attachment delivery, an agent can run cc-connect send --file /absolute/path/to/report.pdf or cc-connect send --image /absolute/path/to/chart.png; Feishu and Telegram are the explicitly documented initial platforms, and attachments default to a 50 MiB limit.
FAQ
Does CC-Connect require a public IP address?
Does it include Codex, Claude Code, or the other coding agents?
How are privileged chat commands restricted?
admin_from = "alice,bob" to allow only those user IDs to run privileged commands such as /dir and /shell. The setting belongs under [[projects]], and users can retrieve their ID with /whoami or /status.Can a long-running shell or test command lock a session indefinitely?
max_turn_time_mins wall-clock limit with soft stop, forced termination, and automatic resume. A user can also interrupt the active turn with /cancel.