Authentication

Every route requires a bearer token:

Authorization: Bearer spor_pat_...

Transport is HTTPS only. Unauthenticated calls are hard-rejected on both the REST and MCP surfaces — there is no anonymous author. Tokens grant full read and write: the trust model is “everyone on the team can read and write the team graph”, the same as a shared repository. The one privileged distinction is the admin gate.

Personal access tokens

Per-user tokens carry the spor_pat_ prefix (legacy sub_pat_ tokens stay valid, no re-mint required). A token’s canonical subject is a person node in the graph, and its {name, email} attribution resolves from that node at read time — an email change re-points the token instead of severing it.

A token may carry an expiry; once past it, the token is rejected like a revoked one.

Three ways to mint one:

• Self-serve: POST /v1/me/tokens mints a token bound to your own person node. See Tokens and agents.
• Admin onboarding: POST /v1/admin/tokens binds a token to someone else’s person node.
• Server box: an operator runs spor-mint-token --person <person-node-id> directly on the server; --expires <N>d|<date> sets an expiry.

Check what a token resolves to with GET /v1/me (Reads). A response with bound: false means the token authenticates but maps to no person node, so routed questions and the personal queue will be empty.

The admin gate

A caller is an admin if and only if their person node carries a stewards edge to the graph root (default org-root). Without it, the admin routes (/v1/admin/*, the ?auth=1 export) return 403 forbidden. The first admin is bootstrapped on the server box with spor-mint-token --admin --person <id>, which writes that stewards edge, creating the person node from --name/--email if needed.

OAuth 2.1 for MCP connectors

Connector hosts (such as claude.ai) cannot carry a static bearer token, so the MCP surface supports the standard OAuth 2.1 discovery chain:

• Protected-resource metadata (RFC 9728), advertised on the /mcp 401 via WWW-Authenticate.
• Authorization-server metadata (RFC 8414).
• Dynamic client registration (RFC 7591).
• Authorization code + PKCE, S256 only, public clients.

The consent step is a PAT exchange: the authorize page asks you to paste your existing spor_pat_ token into the server’s own page. The token never reaches the connector host, so the OAuth identity is exactly the PAT’s {name, email} attribution record.

Token lifetimes:

Token	Prefix	Lifetime
Access token	spor_oat_ (legacy sub_oat_ accepted)	30 days
Refresh token	spor_ort_	90 days, rotating, single-use
Authorization code	—	single-use, 10 minutes

A client holding a refresh token transparently refreshes on a 401/403 (grant_type=refresh_token) and retries once.

CLI sign-in: the device grant

spor auth login (alias spor login) defaults to the OAuth 2.0 device authorization grant (RFC 8628), so it works headless and over SSH:

1. The CLI calls POST /oauth/device_authorization {client_id?, scope?} and receives {device_code, user_code, verification_uri, verification_uri_complete, expires_in, interval}.
2. It prints the URL and code (auto-opening a local browser when one is present); you approve in any browser.
3. It polls POST /oauth/token {grant_type: urn:ietf:params:oauth:grant-type:device_code, device_code} — answering authorization_pending/slow_down until approval — then receives the same person-bound, org-scoped, refreshable spor_oat_/spor_ort_ pair the connector flow mints.

Loopback PKCE

spor auth login --web is the optimization for when a browser runs on the same machine (OAuth 2.1 authorization code + PKCE, RFC 8252). The CLI binds a one-shot http://127.0.0.1:<port>/callback listener, anonymously registers a public client for it (POST /oauth/register, RFC 7591), opens the browser to GET /oauth/authorize with response_type=code, client_id, redirect_uri, code_challenge, code_challenge_method=S256, state, and optional scope, captures the redirected ?code (CSRF-checked against state), and exchanges it at POST /oauth/token {grant_type: authorization_code, code, code_verifier, client_id, redirect_uri} for the same token pair. It then best-effort unregisters the throwaway client (RFC 7592 DELETE), and falls back to the device grant when the server exposes no loopback or registration endpoints.

spor auth login <url> <token> / spor join <url> <token> is the non-interactive paste path; CI stays on SPOR_TOKEN.

Agent session tokens

A person mints a short-lived, per-session token for an agent they own through POST /v1/agents/{id}/token (Tokens and agents). Authorization is ownership (the agent’s owned-by → person edge), never admin, so running your own agents needs no special privilege.

The token’s record carries {agent, session} and no person; the owning person resolves from the owned-by edge at verify time, so a deleted agent or owner makes the token fail closed rather than impersonate. Writes under it are attributed agent on behalf of person: the server stamps authored_by_agent and session alongside the owner as author, and uses authored_via: dispatch — the person → agent chain is the audit trail.

The token can be minted session-deferred (the run session is not known until the run exists) and bound to the real session afterwards via POST /v1/agents/session — write-once, so the binding is always the actual run. Writes before the bind carry no session rather than a phantom one. A long-lived standing variant ({standing: true}) mints a durable agent-scoped spor_pat_ for headless environments.

Render tickets

POST /v1/lens/{id}/ticket mints a signed, expiring, read-only ticket carrying {lens_id, sharer_person_id, exp} — the credential a shared view link carries instead of the sharer’s PAT. It binds the viewer to the recorded sharer (the render shows a “Viewing as” banner), is honored only on GET /v1/lens/{id}/render, and can never authorize a write. Tickets are stateless (HMAC over a server-held key): there is no revocation list; expiry is the bound. See Lenses and sharing.