Crate pmcp_code_mode 

Code Mode - LLM-generated query validation and execution.

This crate provides the infrastructure for “Code Mode”, which allows MCP clients to generate and execute structured queries (GraphQL, SQL, REST) with a validation pipeline that ensures security and provides human-readable explanations.

Architecture

describe_schema() → LLM generates code → validate_code() → user approval → execute_code()

Key Components

• Validation Pipeline: Parse → Policy Check → Security Analysis → Explanation → Token
• Approval Tokens: HMAC-signed tokens binding code hash to validation result
• Explanations: Template-based business-language descriptions of queries
• Policy Evaluation: Pluggable trait for Cedar/AVP/custom policy engines

Example Usage

use pmcp_code_mode::{
    CodeModeConfig, ValidationPipeline, ValidationContext
};

// Create a validation pipeline
let config = CodeModeConfig::enabled();
let pipeline = ValidationPipeline::new(config, b"secret-key".to_vec());

// Validate a query
let context = ValidationContext::new("user-123", "session-456", "schema-hash", "perms-hash");
let result = pipeline.validate_graphql_query("query { users { id name } }", &context)?;

Re-exports

pub use config::resolve_server_id_from_env;

pub use config::CodeModeConfig;

pub use sql::SqlStatementInfo;

pub use sql::SqlStatementType;

pub use sql::SqlValidator;

pub use executor::filter_blocked_fields;

pub use executor::find_blocked_fields_in_output;

pub use executor::ApiCallLog;

pub use executor::ArrayMethodCall;

pub use executor::BinaryOperator;

pub use executor::BuiltinFunction;

pub use executor::CompileError;

pub use executor::ExecutionConfig;

pub use executor::ExecutionPlan;

pub use executor::ExecutionResult;

pub use executor::HttpExecutor;

pub use executor::JsExecutor;

pub use executor::MockExecutionMode;

pub use executor::MockHttpExecutor;

pub use executor::MockedCall;

pub use executor::PathPart;

pub use executor::PathTemplate;

pub use executor::PlanCompiler;

pub use executor::PlanExecutor;

pub use executor::PlanMetadata;

pub use executor::PlanStep;

pub use executor::SdkExecutor;

pub use executor::UnaryOperator;

pub use executor::ValueExpr;

pub use executor::McpExecutor;

pub use validation::ValidationContext;

pub use validation::ValidationPipeline;

pub use templates::TemplateContext;

pub use handler::format_error_response;

pub use handler::format_execution_error;

pub use handler::CodeModeHandler;

pub use handler::CodeModeToolBuilder;

pub use handler::ExecuteCodeInput;

pub use handler::ValidateCodeInput;

pub use handler::ValidationResponse;

pub use policy::get_baseline_policies;

pub use policy::get_code_mode_schema_json;

pub use policy::AuthorizationDecision;

pub use policy::NoopPolicyEvaluator;

pub use policy::OperationEntity;

pub use policy::PolicyEvaluationError;

pub use policy::PolicyEvaluator;

pub use policy::ServerConfigEntity;

pub use policy::get_openapi_baseline_policies;

pub use policy::get_openapi_code_mode_schema_json;

pub use policy::normalize_operation_format;

pub use policy::normalize_path_to_pattern;

pub use policy::OpenAPIServerEntity;

pub use policy::ScriptEntity;

pub use policy::get_sql_baseline_policies;

pub use policy::get_sql_code_mode_schema_json;

pub use policy::SqlServerEntity;

pub use policy::StatementEntity;

pub use policy::cedar::CedarPolicyEvaluator;

pub use avp::AvpClient;

pub use avp::AvpConfig;

pub use avp::AvpError;

pub use avp::AvpPolicyEvaluator;

pub use schema_exposure::CodeModeExposurePolicy;

pub use schema_exposure::DerivationMetadata;

pub use schema_exposure::DerivationStats;

pub use schema_exposure::DerivedSchema;

pub use schema_exposure::ExposureMode;

pub use schema_exposure::FilterReason;

pub use schema_exposure::FilteredOperation;

pub use schema_exposure::GlobalBlocklist;

pub use schema_exposure::McpExposurePolicy;

pub use schema_exposure::MethodExposurePolicy;

pub use schema_exposure::Operation;

pub use schema_exposure::OperationCategory;

pub use schema_exposure::OperationDetails;

pub use schema_exposure::OperationParameter;

pub use schema_exposure::OperationRiskLevel;

pub use schema_exposure::SchemaDeriver;

pub use schema_exposure::SchemaFormat;

pub use schema_exposure::SchemaMetadata;

pub use schema_exposure::SchemaSource;

pub use schema_exposure::ToolExposurePolicy;

pub use schema_exposure::ToolOverride;

Modules

avp

Amazon Verified Permissions (AVP) policy evaluator for Code Mode.

config

Code Mode configuration.

executor

AST-based JavaScript execution for Code Mode.

handler

Code Mode Handler trait for unified soft-disable and tool management.

policy

Policy evaluation framework for Code Mode.

policy_annotations

Policy annotation parser for Cedar policies.

schema_exposure

Schema Exposure Architecture for MCP Built-in Servers.

sql

SQL validation for Code Mode.

templates

Code Mode instruction and policy templates.

validation

Validation pipeline for Code Mode.

Structs

ApiCall

An API call extracted from the JavaScript code.

ApprovalToken

Approval token that authorizes code execution.

CodeLocation

Location in source code.

GraphQLQueryInfo

Information extracted from a parsed GraphQL query.

GraphQLValidator

GraphQL query validator.

HmacTokenGenerator

HMAC-based token generator for MVP.

JavaScriptCodeInfo

Information extracted from parsed JavaScript code.

JavaScriptValidator

JavaScript code validator for OpenAPI Code Mode.

JsCodeExecutor

Adapter bridging [HttpExecutor] to CodeExecutor for JavaScript/OpenAPI servers (Pattern B: JS+HTTP).

McpCodeExecutor

Adapter bridging [McpExecutor] to CodeExecutor for MCP composition servers (Pattern D: JS+MCP).

OutputDeclaration

Declared output type from @returns annotation.

PolicyViolation

A policy violation found during validation.

SafetyViolation

A safety violation found during JavaScript validation.

SdkCodeExecutor

Adapter bridging [SdkExecutor] to CodeExecutor for SDK-backed servers (Pattern C: JS+SDK).

SecurityAnalysis

Security analysis of code.

SecurityIssue

Potential security issues found during analysis.

TemplateExplanationGenerator

Template-based explanation generator for MVP.

TokenSecret

Zeroizing wrapper for HMAC token secrets.

ValidationMetadata

Detailed metadata about a validation.

ValidationResult

Result of validating code through the pipeline.

Enums

CodeLanguage

Supported code languages for validation and execution.

CodeType

Type of code being validated/executed.

Complexity

Estimated complexity of a query.

ExecutionError

Errors that can occur during execution.

GraphQLOperationType

GraphQL operation type.

HttpMethod

HTTP methods that can be called via the api object.

RiskLevel

Risk level assessed for a query or workflow.

SafetyViolationType

Types of safety violations in JavaScript code.

SecurityIssueType

Types of security issues.

TokenError

Errors from token generator construction.

UnifiedAction

Unified action model that maps to business permissions. Works consistently across GraphQL, OpenAPI, and SQL servers.

ValidationError

Errors that can occur during validation.

Traits

CodeExecutor

High-level trait for executing validated code.

ExplanationGenerator

Trait for generating human-readable explanations.

TokenGenerator

Trait for token generators.

Functions

canonicalize_code

Canonicalize code for consistent hashing.

compute_context_hash

Compute a context hash from schema and permissions.

hash_code

Compute the SHA-256 hash of canonicalized code.

Attribute Macros

async_trait