@macaw-ai/client 0.0.0-stage → 0.9.10

package/payload/secureAI/macaw_adapters/mcp/REFERENCE.md

@@ -0,0 +1,460 @@
+# SecureMCP Reference
+
+SecureMCP provides enterprise-grade security for Model Context Protocol through integration with MACAW Control Plane.
+
+**Two main components:**
+1. **SecureMCP** — For MCP servers YOU write (FastMCP-compatible decorators)
+2. **SecureMCPProxy** — For external MCP servers you DON'T control (inline gateway)
+
+---
+
+## SecureMCP (Your MCP Servers)
+
+FastMCP-compatible decorator API with MACAW security. All tool invocations are cryptographically signed, policy-enforced, and audit-logged.
+
+```python
+from macaw_adapters.mcp import SecureMCP, Context
+
+mcp = SecureMCP("calculator")
+
+@mcp.tool(description="Add two numbers")
+def add(a: float, b: float) -> float:
+    return a + b
+
+if __name__ == "__main__":
+    mcp.run()
+```
+
+### Constructor
+
+```python
+SecureMCP(
+    name: str,                              # Server name (required)
+    version: str = "1.0.0",                 # Server version
+    intent_policy: dict = None,             # MAPL policy declaration
+    roots: list[str] = None,                # Filesystem paths server can access
+    **kwargs                                # Additional MACAWClient options
+)
+```
+
+| Parameter | Type | Default | Description |
+|-----------|------|---------|-------------|
+| `name` | str | required | Server name for MACAW registration |
+| `version` | str | `"1.0.0"` | Server version |
+| `intent_policy` | dict | `None` | MAPL policy (resources, denied_resources, etc.) |
+| `roots` | list | `None` | Filesystem paths this server can access (MCP roots) |
+
+### Decorators
+
+#### `@mcp.tool(name, description, prompts)`
+
+Register a tool that can be invoked by clients.
+
+```python
+@mcp.tool(description="Add two numbers")
+def add(a: float, b: float) -> float:
+    return a + b
+
+@mcp.tool(description="Calculate with history")
+def calculate(ctx: Context, op: str, a: float, b: float) -> dict:
+    # ctx parameter gives access to context vault
+    history = ctx.get("calc_history") or []
+    result = a + b if op == "add" else a * b
+    history.append({"op": op, "result": result})
+    ctx.set("calc_history", history)
+    return {"result": result}
+```
+
+| Parameter | Type | Description |
+|-----------|------|-------------|
+| `name` | str | Tool name (defaults to function name) |
+| `description` | str | Tool description (defaults to docstring) |
+| `prompts` | list | Parameter names to treat as AuthenticatedPrompts |
+
+#### `@mcp.resource(uri_pattern, description)`
+
+Register a read-only resource.
+
+```python
+@mcp.resource("calc://history")
+def get_history(ctx: Context) -> dict:
+    return {"history": ctx.get("calc_history") or []}
+
+@mcp.resource("config://{key}")
+def get_config(ctx: Context, key: str) -> dict:
+    return {"key": key, "value": "..."}
+```
+
+#### `@mcp.prompt(name, description)`
+
+Register a prompt template.
+
+```python
+@mcp.prompt(description="Generate a greeting")
+def greeting(name: str) -> str:
+    return f"Hello, {name}! How can I help?"
+```
+
+### Lifecycle
+
+```python
+mcp.run(transport="stdio")    # Blocking run (stdio or sse)
+```
+
+### Properties
+
+| Property | Type | Description |
+|----------|------|-------------|
+| `mcp.name` | str | Server name |
+| `mcp.version` | str | Server version |
+| `mcp.agent_id` | str | MACAW agent ID (after run) |
+| `mcp.roots` | list | Filesystem roots |
+
+---
+
+## Context
+
+Request context passed to tool handlers. Provides access to context vault, progress reporting, sampling, and elicitation.
+
+```python
+@mcp.tool()
+def my_tool(ctx: Context, param: str) -> dict:
+    # Context vault (persistent storage)
+    ctx.set("key", "value")
+    data = ctx.get("key")
+
+    # Logging
+    ctx.info("Processing request")
+    ctx.audit("data_access", target="records", outcome="success")
+
+    return {"result": data}
+```
+
+### Methods
+
+| Method | Description |
+|--------|-------------|
+| `ctx.get(key)` | Get value from context vault |
+| `ctx.set(key, value)` | Set value in context vault |
+| `await ctx.report_progress(progress, message)` | Report progress (0.0-1.0) |
+| `await ctx.read_resource(uri)` | Read another resource |
+| `await ctx.sample(prompt, ...)` | Request LLM completion from client |
+| `await ctx.elicit(prompt, ...)` | Request user input from client |
+
+### Logging Methods
+
+| Method | Description |
+|--------|-------------|
+| `ctx.debug(message)` | Log debug message |
+| `ctx.info(message)` | Log info message |
+| `ctx.warning(message)` | Log warning message |
+| `ctx.error(message)` | Log error message |
+| `ctx.audit(action, target, outcome, **metadata)` | Cryptographically signed audit entry |
+
+### Sampling (MCP Sampling)
+
+Request LLM completion from the calling client:
+
+```python
+@mcp.tool()
+async def analyze(ctx: Context, text: str) -> dict:
+    summary = await ctx.sample(
+        prompt=f"Summarize: {text}",
+        system_prompt="You are a helpful assistant.",
+        max_tokens=500
+    )
+    return {"summary": summary}
+```
+
+**Note on `temperature`:** it is unset by default and is not sent unless you pass
+one. The server does not know which model the client will use, and current Claude
+models reject sampling parameters. Pass a temperature only if you know the client's
+model accepts it — and expect the client may ignore it regardless (MCP treats
+sampling parameters as hints).
+
+### Elicitation (MCP Elicitation)
+
+Request user input during tool execution:
+
+```python
+@mcp.tool()
+async def delete_file(ctx: Context, path: str) -> dict:
+    confirmed = await ctx.elicit(
+        prompt=f"Delete {path}?",
+        input_type="confirm",
+        default="no"
+    )
+    if confirmed:
+        # perform deletion
+        return {"deleted": path}
+    return {"cancelled": True}
+```
+
+---
+
+## SecureMCPProxy (External MCP Servers)
+
+Inline gateway that wraps external MCP servers (Snowflake, Salesforce, Databricks, etc.) with MACAW security.
+
+```python
+from macaw_adapters.mcp import SecureMCPProxy
+
+# HTTP transport (remote MCP servers)
+proxy = SecureMCPProxy(
+    app_name="snowflake-mcp",
+    upstream_url="http://localhost:9000/mcp"
+)
+
+# stdio transport (local MCP servers via subprocess)
+proxy = SecureMCPProxy(
+    app_name="salesforce-dx",
+    command=["npx", "@salesforce/mcp", "--orgs", "DEFAULT"],
+    env={"HOME": os.environ["HOME"]}
+)
+
+# Use tools
+tools = proxy.list_tools()
+result = proxy.call_tool("run_snowflake_query", {"statement": "SELECT 1"})
+```
+
+### Constructor
+
+```python
+SecureMCPProxy(
+    app_name: str,                          # Application name (required)
+    # HTTP transport
+    upstream_url: str = None,               # URL of upstream MCP server
+    upstream_auth: dict = None,             # Auth config (see below)
+    # stdio transport
+    command: list[str] = None,              # Command to start MCP server
+    env: dict = None,                       # Environment variables
+    # Identity
+    iam_token: str = None,                  # IAM token for user identity
+    user_name: str = None,                  # User name
+    intent_policy: dict = None              # MAPL intent policy
+)
+```
+
+| Parameter | Type | Description |
+|-----------|------|-------------|
+| `app_name` | str | Application name for MACAW registration |
+| `upstream_url` | str | URL for HTTP transport |
+| `upstream_auth` | dict | Auth: `{"type": "bearer", "token": "..."}` or `{"type": "api_key", "api_key": "...", "header_name": "X-API-Key"}` |
+| `command` | list | Command for stdio transport |
+| `env` | dict | Environment variables for stdio subprocess |
+| `iam_token` | str | IAM token for authenticated identity |
+| `user_name` | str | User name (used with local identity) |
+| `intent_policy` | dict | MAPL policy for this proxy |
+
+**Note:** Provide either `upstream_url` (HTTP) or `command` (stdio), not both.
+
+### Methods
+
+#### `call_tool(tool_name, params) -> Any`
+
+Call a tool through MACAW security.
+
+```python
+result = proxy.call_tool("run_snowflake_query", {
+    "statement": "SELECT * FROM customers LIMIT 10"
+})
+```
+
+Flow:
+1. MACAWClient.invoke_tool() checks MAPL policy
+2. Signs the invocation cryptographically
+3. Logs to audit trail
+4. Proxy handler forwards to upstream MCP server
+
+#### `list_tools() -> list[dict]`
+
+List available tools discovered from upstream.
+
+```python
+tools = proxy.list_tools()
+# Returns: [{"name": "run_query", "description": "...", "schema": {...}}]
+```
+
+#### `get_tool_schema(tool_name) -> dict`
+
+Get schema for a specific tool.
+
+```python
+schema = proxy.get_tool_schema("run_snowflake_query")
+```
+
+#### `bind_to_user(user_client) -> BoundMCPProxy`
+
+Bind proxy to a specific user's identity for multi-tenant applications.
+
+```python
+from macaw_client import MACAWClient, RemoteIdentityProvider
+
+# Authenticate user via Keycloak/Okta
+jwt, _ = RemoteIdentityProvider().login("alice", "Alice123!")
+alice = MACAWClient(user_name="alice", iam_token=jwt, app_name="my-app")
+alice.register()
+
+# Bind proxy to Alice's identity
+alice_proxy = proxy.bind_to_user(alice)
+
+# Calls now use Alice's identity and policy
+result = alice_proxy.call_tool("run_snowflake_query", {...})
+```
+
+#### `refresh_tools()`
+
+Re-discover tools from upstream (if tools may have changed).
+
+```python
+proxy.refresh_tools()
+```
+
+### Properties
+
+| Property | Type | Description |
+|----------|------|-------------|
+| `proxy.app_name` | str | Application name |
+| `proxy.upstream_url` | str | Upstream URL |
+| `proxy.agent_id` | str | MACAW agent ID |
+| `proxy.server_id` | str | Alias for agent_id |
+| `proxy.tool_schemas` | dict | Discovered tool schemas |
+| `proxy.is_connected` | bool | Connection status |
+
+---
+
+## BoundMCPProxy
+
+MCP Proxy bound to a specific user's identity. All calls use the user's policy and audit trail.
+
+```python
+bound_proxy = proxy.bind_to_user(user_client)
+
+# Methods (same as SecureMCPProxy)
+result = bound_proxy.call_tool("tool_name", params)
+tools = bound_proxy.list_tools()
+schema = bound_proxy.get_tool_schema("tool_name")
+```
+
+---
+
+## Examples
+
+### SecureMCP Calculator Server
+
+```python
+from macaw_adapters.mcp import SecureMCP, Context
+
+mcp = SecureMCP("calculator")
+
+@mcp.tool(description="Add two numbers")
+def add(a: float, b: float) -> float:
+    return a + b
+
+@mcp.tool(description="Multiply two numbers")
+def multiply(a: float, b: float) -> float:
+    return a * b
+
+@mcp.tool(description="Calculate with history")
+def calculate(ctx: Context, op: str, a: float, b: float) -> dict:
+    history = ctx.get("calc_history") or []
+    result = a + b if op == "add" else a * b
+    history.append({"op": op, "a": a, "b": b, "result": result})
+    ctx.set("calc_history", history)
+    return {"result": result, "history_count": len(history)}
+
+@mcp.resource("calc://history")
+def get_history(ctx: Context) -> dict:
+    return {"history": ctx.get("calc_history") or []}
+
+if __name__ == "__main__":
+    mcp.run()
+```
+
+### SecureMCPProxy with Snowflake
+
+```python
+from macaw_adapters.mcp import SecureMCPProxy
+
+proxy = SecureMCPProxy(
+    app_name="snowflake-data",
+    upstream_url="http://localhost:9000/mcp"
+)
+
+# List available tools
+for tool in proxy.list_tools():
+    print(f"  {tool['name']}: {tool['description']}")
+
+# Query data
+result = proxy.call_tool("run_snowflake_query", {
+    "statement": "SELECT CURRENT_USER(), CURRENT_ROLE()"
+})
+print(result)
+```
+
+### Multi-User SecureMCPProxy
+
+```python
+from macaw_adapters.mcp import SecureMCPProxy
+from macaw_client import MACAWClient, RemoteIdentityProvider
+
+# Service creates shared proxy
+proxy = SecureMCPProxy(
+    app_name="data-platform",
+    upstream_url="http://localhost:9000/mcp"
+)
+
+# Authenticate users via Keycloak
+alice_jwt, _ = RemoteIdentityProvider().login("alice", "Alice123!")
+alice = MACAWClient(user_name="alice", iam_token=alice_jwt, app_name="data-platform")
+alice.register()
+
+bob_jwt, _ = RemoteIdentityProvider().login("bob", "Bob@123!")
+bob = MACAWClient(user_name="bob", iam_token=bob_jwt, app_name="data-platform")
+bob.register()
+
+# Per-user proxies
+alice_proxy = proxy.bind_to_user(alice)
+bob_proxy = proxy.bind_to_user(bob)
+
+# Each user's calls use their identity and policy
+alice_result = alice_proxy.call_tool("list_objects", {"object_type": "database"})
+bob_result = bob_proxy.call_tool("run_snowflake_query", {"statement": "SELECT 1"})
+```
+
+---
+
+## Security Model
+
+All security is delegated to MACAW Control Plane:
+
+1. **Registration**: Server/proxy registers with LocalAgent on startup
+2. **Tool Invocation**: Requests routed through ToolAgent PEP
+3. **Policy Enforcement**: MAPL policies enforced on every request
+4. **Cryptographic Signing**: All invocations are signed per-request
+5. **Audit Logging**: Security events logged with cryptographic proof
+6. **Identity Propagation**: User identity flows through bind_to_user()
+
+**Per-invocation security**: Even with persistent connections (stdio), MACAW creates fresh AuthenticatedContext for each call. This solves the "shared context" problem in standard MCP.
+
+---
+
+## Legacy API
+
+The `Server` and `Client` classes are available for backwards compatibility:
+
+```python
+from macaw_adapters.mcp import Server, Client
+```
+
+For new projects, use `SecureMCP` (your servers) or `SecureMCPProxy` (external servers).
+
+---
+
+## Installation
+
+```bash
+pip install macaw-adapters[mcp-proxy]
+```
+
+For more information: https://macawsecurity.ai

package/payload/secureAI/macaw_adapters/mcp/__init__.py

@@ -0,0 +1,65 @@
+"""
+MACAW MCP Adapter - FastMCP-compatible API with MACAW Security.
+
+Two main components:
+1. SecureMCP - For MCP servers YOU write (wrap your tools with MACAW security)
+2. SecureMCPProxy - For MCP servers you DON'T control (inline gateway to external servers)
+
+Usage - SecureMCP (your servers):
+    from macaw_adapters.mcp import SecureMCP, Context
+
+    mcp = SecureMCP("calculator")
+
+    @mcp.tool(description="Add two numbers")
+    def add(a: float, b: float) -> float:
+        return a + b
+
+    mcp.run()
+
+Usage - SecureMCPProxy (external servers):
+    from macaw_adapters.mcp import SecureMCPProxy
+
+    # Connect to external MCP server (Salesforce, Google, etc.)
+    proxy = SecureMCPProxy(
+        app_name="salesforce-mcp",
+        upstream_url="https://mcp.salesforce.com",
+        upstream_auth={"type": "bearer", "token": SF_TOKEN}
+    )
+
+    # Call tools - MACAW security applied
+    result = proxy.call_tool("query_accounts", {"limit": 10})
+
+Features:
+    - FastMCP-compatible decorator API (SecureMCP)
+    - Inline gateway for external MCP servers (SecureMCPProxy)
+    - Policy-enforced tool execution
+    - Cryptographic audit trail
+    - Multi-user identity via bind_to_user()
+
+For more information: https://macawsecurity.ai
+"""
+
+__version__ = "0.9.10.1"
+
+# Primary FastMCP-compatible API (your MCP servers)
+from .mcp import SecureMCP, Context
+
+# Inline Gateway for external MCP servers
+from .proxy import SecureMCPProxy, BoundMCPProxy
+
+# Legacy API (backwards compatibility)
+from .server import Server
+from .client import Client
+
+__all__ = [
+    # Your MCP servers
+    "SecureMCP",
+    "Context",
+    # External MCP servers (inline gateway)
+    "SecureMCPProxy",
+    "BoundMCPProxy",
+    # Legacy
+    "Server",
+    "Client",
+    "__version__"
+]