Skip to content

MCP Integration

Assay is built for the Model Context Protocol.


What is MCP?

Model Context Protocol (MCP) is an open standard for connecting AI agents to external tools and data sources. It defines how agents:

  • Discover available tools (tools/list)
  • Call tools with arguments (tools/call)
  • Receive results

Assay validates these interactions to ensure your agent behaves correctly.


Assay's Role in the MCP Stack

┌─────────────────────────────────────────────────────────────┐
│                        Your Agent                           │
└─────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────┐
│                     MCP (Connectivity)                      │
│              "How agents talk to tools"                     │
└─────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────┐
│                   Assay (Quality Engineering)               │
│        "Are those conversations correct, safe, repeatable?" │
└─────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────┐
│                      External Tools                         │
│              Databases, APIs, File Systems                  │
└─────────────────────────────────────────────────────────────┘

MCP without Assay = unverified traffic.


Integration Patterns

Assay integrates with MCP in three ways:

1. Trace Consumer (Offline Testing)

Import MCP sessions and run deterministic tests in CI.

assay import --format inspector session.json
assay run --config eval.yaml --strict

Use case: CI regression gates, debugging, baseline comparison.

Quick Start


2. MCP Wrapper (Runtime Validation)

Expose Assay as MCP tools that agents call before executing actions.

assay mcp wrap --policy assay.yaml -- <real-mcp-command> [args...]

The agent can query: - assay_check_args — "Is this argument valid?" - assay_check_sequence — "Is this call order allowed?" - assay_policy_decide — "Should I proceed?"

Use case: Agent self-correction, runtime guardrails.

Server Guide


3. MCP Gateway (Enterprise)

Inline enforcement for production deployments.

Agent ──► Assay Gateway ──► MCP Server ──► Tools
              └─► Capture, Redact, Enforce, Sign

Use case: Compliance logging, policy enforcement, audit trails.

Gateway Guide (Enterprise)


What Assay Validates

MCP standardizes how agents communicate. Assay validates what they communicate.

Validation Question Metric
Argument Correctness Are tool arguments schema-valid? args_valid
Sequence Validity Are calls in the right order? sequence_valid
Blocklist Enforcement Was a forbidden tool called? tool_blocklist
Replay Fidelity Can we reproduce this incident? replay

Supported Formats

Import Formats

Format Source Command
MCP Inspector MCP Inspector --format inspector
JSON-RPC 2.0 Raw MCP messages --format jsonrpc
LangChain LangChain traces Not yet supported in assay import
LlamaIndex LlamaIndex traces Not yet supported in assay import

Export Formats

Format Use Case Flag
SARIF GitHub Code Scanning assay ci --sarif <path>
JUnit CI test results assay ci --junit <path>
JSON Programmatic access assay validate --format json --output <path>

Quick Comparison

Feature MCP Alone MCP + Assay
Tool discovery
Tool execution
Argument validation
Sequence enforcement
Blocklist
Deterministic replay
CI integration
Offline testing

Next Steps

  • Quick Start


    Import your first MCP session in 5 minutes.

    Quick Start

  • Assay MCP Wrapper


    Let agents validate their own actions.

    Server Guide

  • Self-Correction


    Build agents that fix their own mistakes.

    Self-Correction

  • Import Formats


    Supported log formats and conversion.

    Import Formats