ValueSignal connector
For Claude, Claude Code, Codex and ChatGPT. Last updated: November 1, 2026.
What the connector does
ValueSignal keeps a private, human-in-the-loop record of how you work with AI, and lets you turn that record into verifiable proof you control. The connector is a hosted Model Context Protocol (MCP) server that AI apps call on your behalf, with your explicit permission. It does nothing on its own: your AI app calls a tool only when you ask it to, for example "log this exchange to ValueSignal".
Server URL: https://app.valuesignal.ai/mcp
The connector exposes four tools. Two only read, two write to your own account.
| Tool | What it does | Reads or writes | Permission |
|---|---|---|---|
| Check ValueSignal connection | Reports which ValueSignal account is connected (email and id), which app is connected, whether that app is a verified first-party client, and which permissions were granted. | Read only | profile:read |
| Open the ValueSignal dashboard | Returns the URL of your dashboard so the app can open it. Makes no network request. | Read only | none |
| Log an AI turn to ValueSignal | Sends one exchange (your prompt and the assistant's response) to your private logbook for scoring. Repeating the same turn returns the original result instead of creating a duplicate. | Writes to your logbook | capture:write |
| Mint a Proof of Work certification | Creates a verifiable certification from work already in your logbook and returns the files to commit into a repository plus a public verify URL. Minting the same scope twice in a day returns the same certification. | Writes to your account; the verify URL is public | proof:write |
No tool deletes anything. Revoking a certification is a separate action in the ValueSignal web app.
Connect it from your app
You need a ValueSignal account first: create one or sign in. Every connection goes through the same sign-in and consent page, where you see the app's name, the address it returns to, and the permissions it asks for, and choose Allow or Deny.
Claude (web, Desktop, mobile, Cowork)
- Open Claude's Settings → Connectors. If ValueSignal is listed in the directory, choose it; otherwise choose Add → Add custom connector, name it
ValueSignal, and enter the server URLhttps://app.valuesignal.ai/mcp. Leave the client id and secret fields empty. - Choose Connect, sign in to ValueSignal, review the permissions, and choose Allow.
- In a new chat, ask Claude to check your ValueSignal connection. The first time a tool runs, Claude asks you to approve it.
Claude Code
claude mcp add --transport http valuesignal https://app.valuesignal.ai/mcp
Then, inside a Claude Code session, run /mcp, select valuesignal, and choose Authenticate. Your browser opens the sign-in and consent page and returns to Claude Code when you allow it.
Codex (desktop app)
- Open Settings → MCP servers → Add server.
- Choose the Streamable HTTP transport, name it
valuesignal, and enterhttps://app.valuesignal.ai/mcp. Leave the authentication method on automatic and do not enter a bearer token. - Choose Authenticate, sign in, and allow the permissions. Start a new chat afterwards: a chat opened before the server was added does not see its tools.
ChatGPT
Install ValueSignal from the ChatGPT directory once it is listed. Connecting works the same way: sign in to ValueSignal and allow the permissions.
Permissions and consent
Consent is explicit and shown on every new connection, even if you connected the same app before. The connector asks for these permissions and no others:
| Permission | What it allows |
|---|---|
profile:read | Read your account email and id to confirm which account is connected. |
capture:write | Log AI turns you ask it to capture into your private ValueSignal logbook. |
proof:write | Mint Proof of Work certifications from your captured work. |
offline_access | Stay connected without signing in again, until you disconnect it. |
Each app holds its own permissions. You can see every connected app, when it last used the connection, and disconnect any of them at Account Settings → Connected apps. Disconnecting takes effect on the app's next request. Access tokens last one hour and are renewed automatically; a connection that goes unused for 30 days expires on its own.
What happens to your data
- Only what you ask to capture is sent. The connector has no access to your chat history, memory, files, or other conversations. Your AI app sends an exchange only when you ask it to log that exchange.
- Captured turns are private. They are stored in your own logbook, visible only to you, and scored into skill signals. Credentials and secrets that appear in a turn are redacted on the server before anything is stored.
- Proofs are public only when you mint them. A certification's verify URL shows the certification, not your logbook. You choose when to mint one and where to share it.
- The app's identity is recorded, not trusted. Each capture records which app sent it and whether that app is a verified first-party client. A repository name an app declares is recorded as declared by the app and never used on its own to certify project work.
- Deletion. Delete captured turns from your logbook, disconnect an app from Account Settings, or delete your account. The privacy policy covers retention and your rights.
For IT administrators
Every request from an AI app is authenticated with a short-lived OAuth 2.0 access token issued to a user who signed in and consented. There are no API keys, shared secrets or custom headers, and the server never accepts a credential in a URL.
| Endpoint | URL |
|---|---|
| MCP server (Streamable HTTP, POST) | https://app.valuesignal.ai/mcp |
| Protected resource metadata (RFC 9728) | https://app.valuesignal.ai/.well-known/oauth-protected-resource/mcp |
| Authorization server metadata (RFC 8414) | https://app.valuesignal.ai/.well-known/oauth-authorization-server/oauth |
| Authorization, token, registration, revocation | https://app.valuesignal.ai/oauth/authorize, /oauth/token, /oauth/register, /oauth/revoke |
- Protocol. OAuth 2.0 authorization code with PKCE (S256 only), as the MCP authorization specification requires. The authorization server issues opaque access tokens valid for one hour, bound to the MCP resource, and rotating refresh tokens with a 30-day idle expiry. Reuse of a rotated refresh token revokes the connection.
- Client registration. Apps identify themselves through Dynamic Client Registration (RFC 7591) or a Client ID Metadata Document. Registration accepts
httpsredirect URIs and loopback addresses for native apps such as Claude Code and Codex, nothing else. Self-registered apps are labelled unverified on every record they create; first-party apps are verified by configuration on the server, never by anything the app sends. - Consent. Shown on every authorization, after the user signs in to ValueSignal with their existing credentials (two-factor authentication where the user has enabled it).
- Transport. HTTPS only. Unauthenticated requests receive
401with aWWW-Authenticatechallenge that points at the protected resource metadata. - Rate limits. Registration 20 per hour per address, token requests 60 per minute per address, tool calls 120 per minute per user.
- Scope of access. A token can only write to the logbook of the user who consented and read that user's own profile. The connector cannot read other users' data, organization data, or anything in the AI app.
- Operator. ValueShift LLC. See the privacy policy and terms of service.
Troubleshooting
- "Authorization with the MCP server failed" or the consent page says the session was not found. The consent page is valid for 30 minutes. Start the connection again from the app and allow it in one pass.
- The app says it is not authenticated after you connected. Chats opened before the connector was added keep their old tool list. Start a new chat and ask it to check the ValueSignal connection.
- Two ValueSignal tool sets appear. You also have the local ValueSignal plugin installed. Both work; name the connector when you ask, or remove the plugin.
- The app asks you to reconnect. The connection was disconnected in Account Settings, expired after 30 days without use, or was revoked after a refresh token was reused. Connect again from the app.
Support
Email support@valuesignal.ai. Include the app you are connecting from and the time of the attempt; never include tokens or passwords.