--- name: signavos-esign description: Send documents for signature through Signavos. Use when the user wants a PDF signed, wants to check whether a document has been signed yet, or wants the signed file and its certificate. Sending costs one credit and must be confirmed. --- # Signavos — e-signature for agents Signavos is a pay-per-signature e-signature service. You drive it through an MCP server, so you can take a PDF from upload all the way to a signed, audit-trailed document without the user touching the web app. **Endpoint:** `https://signavos.com/api/mcp` (MCP Streamable HTTP) **Auth:** OAuth 2.1 with PKCE and dynamic client registration. There are no API keys. The user approves your access on a consent screen and can revoke it at any time. ## Read this before your first call 1. **Call `get_account` first.** It returns the credit balance and every limit you must work within. Do not guess a limit — read it. 2. **Sending spends the user's money.** One credit per document. Always tell the user what you are about to send and to whom, and only then confirm the charge. 3. **You cannot buy credits.** If the balance is too low, say so and point the user at the dashboard. Never look for another way to pay. 4. **You cannot sign anything.** Recipients sign themselves from their own email link. If the user asks you to sign for them or for someone else, explain that no such tool exists. ## Workflow 1. **Upload** — Hand the agent a PDF. It validates, scans and stores it exactly as the web app does. 2. **Prepare** — The agent adds recipients and places signature fields — by preset, so it never has to guess coordinates. 3. **Confirm** — Sending spends 1 credit, so the agent must confirm the charge explicitly before anything is sent. 4. **Send** — Recipients get an email link and sign in their browser. No account, no password. 5. **Track** — The agent polls status and tells you the moment the document is signed. 6. **Collect** — It fetches the signed PDF and the signing certificate with the audit trail. ## Tools available now | Tool | What it does | |---|---| | `get_account` | Read the account email, credit balance and every limit below. | | `list_documents` | List documents with their status, filtered and paged. | | `get_document` | Inspect one document: recipients, fields, artifacts and what to do next. | | `get_download_url` | Get a time-limited link to the signed PDF or its signing certificate. | | `upload_document` | Upload a PDF inline, up to 3 MB. | | `create_upload_url` | Get a signed URL for a larger PDF, up to 40 MB. | | `complete_upload` | Finish a signed-URL upload; the file is validated and scanned like any other. | | `add_recipient` | Add a signer by name and email. | | `remove_recipient` | Remove a signer from a draft. | | `place_fields` | Place signature, initials, name, date, text and checkbox fields — by preset or exact position. | | `send_document` | Send for signature. Spends 1 credit and requires explicit confirmation. | ## Limits | Limit | Value | |---|---| | Credits per document sent | 1 | | Recipients per document | up to 20 | | Inline upload (base64) | 3 MB max | | Upload via signed URL | 40 MB max | | Drafts kept per account | 2 | ## Rules you must not break - **You approve it, and you can revoke it.** Agents connect through an OAuth consent screen you see and approve. Remove the agent from your authorized applications and access stops immediately. - **An agent can never spend money.** There is no purchase tool. Credits are only ever bought by you in a browser — an agent can spend the credits you already own, one per document, and nothing else. - **Sending always needs explicit confirmation.** The send tool refuses to run unless the agent confirms the charge on purpose, so an exploratory prompt cannot mail a contract by accident. - **It only ever sees your account.** Every request is scoped to your account on the server. An agent cannot name someone else's account, and it never receives an account identifier to tamper with. - **The same audit trail as the web app.** Agent actions are recorded in the same tamper-evident chain as your own, tagged with the agent that performed them — so the signing certificate stays legally defensible. - **Signers stay human.** Signavos has no tool that signs on someone's behalf. Recipients always sign the document themselves, from their own email link. ## Handling errors Every failure comes back with a stable `code`, a human-readable `message`, and often a `hint` telling you what to do next. Read the `code`, not the prose. The ones worth handling explicitly: - `INSUFFICIENT_CREDITS` — stop and tell the user to buy credits in the dashboard. Do not retry. - `CONFIRMATION_REQUIRED` — you tried to send without confirming the charge. Ask the user, then retry with the confirmation. - `PAYLOAD_TOO_LARGE` — the file is above the inline limit; use the signed-URL upload instead. - `DOCUMENT_NOT_SENDABLE_*` / `ALREADY_SENT` — the document is past the point of sending. Check its status before deciding what to do. - `RATE_LIMITED` — wait for the window in `details.retry_after_sec` before retrying. - `NOT_FOUND` — the document does not exist **or** does not belong to this user. Do not probe for other identifiers. Anything with `retryable: false` will fail identically if you retry it. Report it and stop. ## Connecting **Claude Code** ```bash claude mcp add --transport http signavos https://signavos.com/api/mcp ``` - Then run /mcp, choose signavos, and approve the consent screen. **Claude (desktop & web)** - Open Settings → Connectors → Add custom connector. - Paste https://signavos.com/api/mcp as the server URL, then Add and Connect. - Approve the consent screen that opens in your browser. **Cursor** - Command Palette → Cursor Settings → Tools & MCP → New MCP Server. - Add: { "mcpServers": { "signavos": { "url": "https://signavos.com/api/mcp" } } } - Select "Needs login" and authorize. **Any MCP client** - Streamable HTTP endpoint: https://signavos.com/api/mcp - Authorization is OAuth 2.1 with PKCE and dynamic client registration — discovery starts at /.well-known/oauth-protected-resource.