Reference
Tools for coding agents.
The CLI's MCP tools, with their input schemas and examples.
CLI 0.2.11
MCP tools
Run cscale mcp over standard input and output. These tools are included by Free or Pro. A connected session discovers the tools its account can use.
- MCP contact_support
Send the supplied message, CLI version, OS and architecture to support. No code, diffs, excerpts, paths or environment values are attached. A person replies by email to the account owner. Uses a personal sign-in, agent token or check-only key on every plan. - MCP get_namespace_plan
Read your namespace's effective plan and public feature flags. Needs the person's sign-in; a check-only key cannot call it. - MCP key_list
List check-only key metadata using the person's personal sign-in. Never uses a check-only key. - MCP key_rename
Rename a check-only key using the person's personal sign-in. - MCP key_revoke
Revoke a check-only key immediately using the person's personal sign-in. - MCP signup_start
Start sign-up for a new account. Ask the person to open the email link and press Confirm, then call signup_wait. Returns a local attempt ID; polling codes and keys stay private. For an existing account, ask the person to run `cscale auth login --email you@example.com`, open the sign-in link in their email and type the code shown in their terminal. - MCP signup_wait
Resume sign-up. Store the issued key privately and return its agent identity, never the key. A pending result includes the next polling delay. Repeating a completed attempt does not collect another key. - MCP tests_accept
Accept a finding from the newest matching saved full check with a reason, recorded as the agent's. Acceptance reruns nothing; the next step names tests_check. On Trusted paths with the passkey layer off, an unlisted agent counts and is flagged. Guarded and Manual paths require a target-listed agent key or a person. When the layer is on or unknown, asks the person to confirm in the CLI. - MCP tests_check
Run the local test check for committed inputs. The version-1 report or problem is in structuredContent, with a summary in text. A passing check can still have findings to review. Recognised comment-only and docstring-only hunks may be left out. Without an account, the check does not run and reports failure. A focused recheck uses installed machine state unless isolated. Use tests_accept to record an agent acceptance when the policy allows. - MCP tests_consent
Read the person's training consent and wording versions using their personal sign-in. Check-only keys cannot read consent. This tool never opts in. - MCP tests_consent_withdraw
Withdraw training consent and delete kept excerpts. Uses a check-only key or personal sign-in. Never opts in. - MCP tests_usage
Read this UTC month's model check count, cap and reset time. External check keys need their owner's namespace; saved sign-up keys already know it. - MCP whoami
Show the account and the agent that tests_check uses. Works without a sign-in. The result is in structuredContent and text.
Remote server
The remote server is https://commitscale.com/mcp. In claude.ai, add a custom connector with this URL. In ChatGPT, add a custom app with this URL. Sign in and approve the connection on the first tool call.
In Claude Code, run claude mcp add --transport http commitscale https://commitscale.com/mcp.
A connection can request read and write. Where your plan offers decisions, a write connection can vote as you; its votes name the app, and in_person decisions refuse it. It can approve deployments as a named reviewer and rerun deployment or release work. An environment can refuse connected apps for deployment approvals and reruns. An app rerun already queued stays queued and can still run under the usual admission rules. People and agents can rerun within their usual limits. Read how connected apps decide. It cannot administer the server, create credentials, or change how you sign in. End a connection in your account's Security settings. When your plan adds tools, reconnect to see them.
Requests are rate-limited; after a 429, wait the seconds in Retry-After. Local test checks and sign-up use the CLI or local MCP server. The remote server does not offer wait_for_change.