Integration Guide

Logito + Claude Code

Claude Code can write code, deploy services, and modify configuration. Logito tells it whether those changes actually worked at runtime. Together, they create a closed-loop development workflow where every change is validated against actual system behavior.

Why this matters

AI agents need runtime feedback

When Claude Code modifies your API handler or updates a database migration, it has no way to verify whether the change worked correctly at runtime. Unit tests validate code correctness. But they don't tell you whether system behavior changed in ways that matter — latency regressions, new error codes, missing response fields, or cascading failures.

Logito closes this gap. It captures actual runtime behavior, compares it against a known baseline, and produces a structured verdict that Claude Code can read and act on.

Setup

Step 1: Install Logito CLI

brew tap progadigital/tap
brew install logito

No account required. The CLI works locally with no cloud dependency.

Step 2: Install the MCP bridge

logito codex install

This installs the Logito MCP bridge, which exposes runtime intelligence tools to Claude Code. The bridge reads your CLI profile and authenticates automatically.

Step 3: Use in Claude Code

Once installed, Claude Code can use Logito's MCP tools directly:

# Claude Code can now use these tools:

logito.get_latest_run          # Get the most recent run summary
review_run                     # Review a run for drift
logito.get_run_diff            # Get structured diff vs baseline
logito.explain_issue           # Investigate a specific issue
logito.get_recommended_action  # Get the next recommended action
logito.get_system_status       # Check overall system health

Workflow

The closed-loop pattern

Claude Code makes a change

Modifies an API handler, updates a migration, changes configuration.

Tests run and Logito captures behavior

During the test run, Logito records HTTP requests/responses, latency, error rates, and response shapes.

Claude Code checks the verdict

Calls review_run to get a structured assessment: PASS (no drift), WARNING (minor changes), or FAIL (regression detected).

If regression: investigate and fix

Claude Code calls logito.explain_issue to understand the specific regression, then modifies the code to fix it.

Re-run and verify

The cycle repeats until the review returns PASS. The agent has validated its own work against actual runtime behavior.

Example

Claude Code fixing a regression

# 1. Start capture
logito dev start --project my-api

# 2. Run tests (Claude Code does this after making changes)
npm test

# 3. Claude Code reviews via MCP
> review_run
  Status: FAIL
  Regression: POST /checkout latency increased 120ms -> 450ms
  Confidence: high (8 endpoints verified, 3 observations on weakest edge)

# 4. Claude Code investigates
> logito.explain_issue { "service": "api" }
  Driver: New database query in checkout handler missing index
  Evidence: 3 slow queries > 300ms in the last 2 minutes

# 5. Claude Code fixes and re-runs
# ... adds database index ...
npm test

# 6. Review passes
> review_run
  Status: PASS
  All 8 endpoints verified stable vs baseline.

Advanced

Local sidecar for air-gapped environments

For environments where the CLI can't reach the cloud API, use the self-hosted sidecar:

docker run -p 3000:3000 -v logito-data:/data logito-local

# The CLI and MCP bridge auto-detect the sidecar at localhost:3000

All analysis runs locally with deterministic-only intelligence. No cloud dependency, no data leaves the network.

Schema stability

MCP tool contract guarantee

Logito's MCP tools follow semantic versioning. The current contract version is 1.1.0 with stability level stable. This means:

Tool names and argument schemas will not change within version 1.x
New tools and optional arguments may be added in minor versions
Breaking changes require a major version bump with 90-day migration period
The contract_version and contract_stability fields are included in every tools/list response

Install

Start validating Claude Code changes against runtime behavior.

Install CLI MCP Tools Reference