SignDocs Brasil MCP Server
Enables AI clients to create and manage e-signature sessions, envelopes, documents, and webhooks through the SignDocs Brasil API.
README
SignDocs Brasil — MCP Server
A Model Context Protocol server for the SignDocs Brasil e-signature API. It lets MCP-capable AI clients (Claude Desktop, Claude Code, Cursor, …) create signing sessions, manage multi-signer envelopes, upload/download documents, verify signatures, and manage webhooks — the same action catalog as the official n8n, Zapier, and Make.com integrations.
It is a thin adapter over the official @signdocs-brasil/api
SDK, which owns OAuth2 token exchange, caching, retries, and error handling.
Install
npm install -g @signdocs-brasil/mcp-server # or run on demand with npx
Credentials
Create an API credential in the SignDocs dashboard (app.signdocs.com.br → API) and expose it as environment variables:
| Variable | Required | Default | Notes |
|---|---|---|---|
SIGNDOCS_CLIENT_ID |
yes | — | OAuth2 client id |
SIGNDOCS_CLIENT_SECRET |
yes | — | OAuth2 client secret |
SIGNDOCS_ENVIRONMENT |
no | hml |
hml (staging) or production |
SIGNDOCS_BASE_URL |
no | derived | override the resolved base URL |
SIGNDOCS_SCOPES |
no | full set | space-separated scope override |
Start in
hml. HML data expires after ~7 days and is safe for testing. Switch toproductiononly when you intend to create real, legally-binding signatures.
Connect an AI client
Claude Desktop (claude_desktop_config.json):
{
"mcpServers": {
"signdocs": {
"command": "npx",
"args": ["-y", "@signdocs-brasil/mcp-server"],
"env": {
"SIGNDOCS_CLIENT_ID": "your_client_id",
"SIGNDOCS_CLIENT_SECRET": "your_client_secret",
"SIGNDOCS_ENVIRONMENT": "hml"
}
}
}
}
Claude Code:
claude mcp add signdocs \
-e SIGNDOCS_CLIENT_ID=your_client_id \
-e SIGNDOCS_CLIENT_SECRET=your_client_secret \
-e SIGNDOCS_ENVIRONMENT=hml \
-- npx -y @signdocs-brasil/mcp-server
Tools
| Tool | Action | Safety |
|---|---|---|
create_signing_session |
Create single-signer session, returns signingUrl |
⚠️ binding + quota |
get_signing_session_status |
Poll session status | read |
get_signing_session |
Full session bootstrap | read |
list_signing_sessions |
List by status | read |
cancel_signing_session |
Cancel a session | ⚠️ irreversible |
resend_signing_session_otp |
Resend OTP | write |
create_envelope |
Multi-signer envelope | ⚠️ binding + quota |
get_envelope |
Envelope details | read |
add_session_to_envelope |
Add a signer, returns signingUrl |
⚠️ binding + quota |
get_envelope_combined_stamp |
Combined stamped PDF URL | read |
upload_document |
Attach a PDF to a transaction | write |
download_document |
Presigned download URLs | read |
list_transactions |
Search/list transactions | read |
get_transaction |
Transaction details | read |
cancel_transaction |
Cancel a transaction | ⚠️ irreversible |
get_evidence |
Cryptographic evidence | read |
verify_evidence |
Public evidence verification | read |
verify_envelope |
Public envelope verification | read |
verify_document |
Detect signatures in a PDF | ⚠️ PROD-only + quota |
register_webhook / list_webhooks / delete_webhook / test_webhook |
Webhook management | mixed |
⚠️ tools carry destructiveHint annotations and a warning in their
description so compliant clients prompt the human before invoking them.
Annotations are only hints — review your client's auto-approval settings.
Not yet exposed
Trust sessions (/v1/trust-sessions) and resend-invite are not in
@signdocs-brasil/api v1.6.1 yet; they'll be added when the SDK supports them.
Digital ICP-Brasil A1 signing runs through the lower-level transaction/advance
flow rather than a hosted-session profile.
Resources
The server exposes grounding resources the model can read on demand:
signdocs://quickstart— the minimal signing flow + safety notessigndocs://policy-profiles— validpolicyProfilevalues and CUSTOM stepssigndocs://webhook-events— all subscribable event types
Remote HTTP transport (multi-tenant)
The same tools are also served over Streamable HTTP so a single deployment can serve many AI agents/tenants — each authenticates per session with its own SignDocs credentials (no shared secret baked into the server).
npm run start:http # or: signdocs-mcp-http (listens on PORT, default 3000)
# or containerized:
docker build -t signdocs-mcp . && docker run -p 3000:3000 signdocs-mcp
Endpoint: POST /mcp (Streamable HTTP). Auth is required on the MCP
initialize request, via the Authorization header:
Authorization: Bearer <token>— a SignDocs OAuth2 access token (from/oauth2/token), passed straight through to the API.Authorization: Basic base64(clientId:clientSecret)— the server runs theclient_credentialsexchange for you.X-SignDocs-Client-Id+X-SignDocs-Client-Secret— the same client credentials as two plain headers (no base64), for header-only clients that can't transform values.
Pick the environment per session with X-SignDocs-Environment: hml|production
(defaults to the server's configured default).
The server behaves as an OAuth 2.0 Resource Server: it serves
GET /.well-known/oauth-protected-resource (RFC 9728, pointing at the SignDocs
authorization server) and answers an unauthenticated initialize with 401 +
WWW-Authenticate. The SignDocs API remains the authoritative token validator.
GET /healthz is an unauthenticated health probe.
Example client config (Bearer):
{
"mcpServers": {
"signdocs-remote": {
"type": "http",
"url": "https://your-host.example/mcp",
"headers": {
"Authorization": "Bearer <signdocs_access_token>",
"X-SignDocs-Environment": "hml"
}
}
}
}
Server env vars: PORT, HOST, SIGNDOCS_ENVIRONMENT (default env),
MCP_PUBLIC_URL (for resource metadata behind a proxy), MCP_CORS_ORIGIN,
MCP_DNS_REBINDING_PROTECTION=true + MCP_ALLOWED_HOSTS / MCP_ALLOWED_ORIGINS
(recommended in production).
Sessions are held in process memory, so run a single instance or use sticky routing. For multi-instance/serverless, front it with sticky sessions or swap the session map for a shared store + EventStore (resumability). Deploying onto the existing
external-apiLambda + API Gateway as a NestedStack is the intended production path.
AWS Lambda
For serverless hosting, @signdocs-brasil/mcp-server/lambda exports
createLambdaHandler — an API Gateway HTTP API v2 handler that runs the MCP
transport statelessly (one server per invocation, no session store), with the
same Bearer/Basic auth. SignDocs hosts this on mcp-hml.signdocs.com.br /
mcp.signdocs.com.br.
import { createLambdaHandler } from '@signdocs-brasil/mcp-server/lambda';
export const handler = createLambdaHandler({ defaultEnvironment: 'hml' });
Development
npm install
npm run build # tsc → dist/
npm test # vitest (pure unit tests, no network)
npm run inspect # build + launch MCP Inspector against the stdio server
Roadmap
- v0.1: local stdio server, full tool catalog, env credentials.
- v0.2 (this release): remote Streamable-HTTP transport with per-session, per-tenant auth (Bearer passthrough or Basic client-credentials) and OAuth Resource Server discovery. Tool layer is shared between both transports.
- Next: deploy the HTTP transport onto
external-api(Lambda + API Gateway NestedStack); optional edge JWT validation + shared-store sessions for horizontal scale.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.