DocsDeveloper integration

MCP tools, configuration and validation

Document configuration, stdio/HTTP connections, tool schemas and authorization.

What MCP provides

MCP discovers and invokes external tools and resources; it does not replace identity, application authorization or approval. Client source includes stdio and HTTP transports. Tool descriptions, read-only annotations and returned text come from the server and do not grant execution rights. Confirm protocol version and capabilities during handshake and integration testing.

Local stdio configuration

This shape matches the current parser. The command and script are placeholders: install and review the real server first. command is an executable path and args is an argument array, not a concatenated shell command. Use minimal directory access for the runtime identity.

json
{
  "transport": "stdio",
  "command": "/absolute/path/to/node",
  "args": [
    "/absolute/path/to/read-only-connector.mjs"
  ],
  "env": {}
}

Remote HTTP configuration

The example domain provides no service. Current configuration requires HTTPS remotely, allowing HTTP only on loopback. URLs reject embedded credentials, queries and fragments; use controlled headers for credentials. Placeholders are not expanded from environment variables. Provision and rotate real secrets through an approved administrative process.

json
{
  "transport": "http",
  "url": "https://connector.example.invalid/mcp",
  "headers": {
    "Authorization": "Bearer REPLACE_WITH_SCOPED_TOKEN"
  }
}

Current configuration limits

The parser also accepts a mcpServers wrapper containing exactly one server. It rejects unknown fields and caps the configuration at 64,000 bytes, args at 64 items and env or headers at 32 entries. Do not import another client's full configuration without checking extra options.

Tool schemas and results

Describe tool inputs with a name, description and JSON Schema, bounding fields and volume. Return structured results or explicit errors, with source, capture time and completeness. This is a read-only tool a third-party server could implement, not an available built-in tool. readOnlyHint is a hint, not permission enforcement.

json
{
  "name": "lookup_order",
  "description": "Read one order from the authorized store.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "orderId": {
        "type": "string",
        "minLength": 1,
        "maxLength": 120
      }
    },
    "required": [
      "orderId"
    ],
    "additionalProperties": false
  },
  "annotations": {
    "readOnlyHint": true
  }
}

Validate discovery and execution

Check startup, handshake, discovery, non-sensitive sample calls, denied access, reconnect and revocation. Record server/client versions, tool name, redacted inputs and evidence. Configuration state and tool counts show discovery, not successful execution.

All documentationNext guide

Still have a question?

Explore expert support