Developer integration routes¶

TrustWeave works from local declarations and already-recorded metadata. Choose the route that matches the files you already have; every command below reads a local input and writes local deterministic evidence. None installs a framework, imports application code, executes an agent or tool, connects to an MCP server, calls a model, reads credentials, or publishes an artifact.

Choose a local input¶

You already have	Start with	What TrustWeave produces	What it does not establish
An Agent Security Manifest and policy	`scan` and `test`	A bundle of declared-flow decisions and synthetic policy results	Runtime behavior, deployment approval, or an agent-security verdict
A LangGraph `langgraph.json` descriptor	`framework-import --framework langgraph`	A normalized local framework inventory	Graph import, compilation, execution, or environment loading
An exported OpenAI Agents descriptor	`framework-import --framework openai-agents`	A normalized local framework inventory	SDK installation, model calls, tool calls, or MCP access
A JSON-compatible CrewAI agents/tasks snapshot	`framework-import --framework crewai`	A normalized local framework inventory	Crew/task execution, guardrail evaluation, or tool resolution
A saved MCP `tools/list` response	`mcp-import`	A stable local MCP tool inventory	Server discovery, connection, authentication, or tool invocation
A repository CI job	The local CI workflow	Deterministic evidence generated inside your chosen runner	Artifact publishing, SARIF upload, or an automatic merge/deployment decision

Copy-paste local examples¶

Run these commands from a TrustWeave source checkout after completing the installation and five-minute local review.

# Read a checked-in LangGraph-style descriptor without importing Python code.
trustweave framework-import \
  --framework langgraph \
  --input examples/frameworks/langgraph-minimal-project/langgraph.json \
  --output-dir artifacts/langgraph-minimal

# Normalize a locally exported OpenAI Agents descriptor without installing its SDK.
trustweave framework-import \
  --framework openai-agents \
  --input examples/frameworks/openai-agents-descriptor.json \
  --output-dir artifacts/openai-agents

# Normalize a saved MCP tools/list snapshot without contacting an MCP server.
trustweave mcp-import \
  --tool-list examples/mcp-tools/support-tools-list.json \
  --output-dir artifacts/mcp-inventory

Use python -m trustweave in place of trustweave when you prefer a Python module invocation. Both entry points expose the same command surface.

Continue from the inventory¶

A framework or MCP inventory is review input, not an Agent Security Manifest. A human reviewer must still classify tools, define sources and flows, select deterministic policy, and run a normal local review. Continue with the manifest contract, policy review, and local CI workflow.

For accepted input shapes, output contracts, and framework-specific limits, see the source guides: framework declaration import and local MCP tools-list import.