Trust Engine AI: model retirements and agent cost

trust-engine-mcp

Free MCP server from Trust Engine AI. It answers questions agents otherwise search the web for:

• Which AI models does this code use, and are any being retired? scan_for_models finds the model ids in the working directory (read-only, on your machine) and checks them against the calendar. The same scan runs as a command (trust-engine-mcp check) and as a GitHub Action, without an agent.
• Is a model we use being retired, and when? Verified dates for Anthropic, OpenAI, Google Gemini API and Vertex AI, AWS Bedrock and Mistral. Each date has its official source URL and the date it was checked.
• Why do our agents cost so much? It reads an OpenAI or Anthropic usage/cost export locally and shows spend by model, cost drivers and the fixes you can make yourself.
• Is a cheaper model good enough? You get our public benchmark numbers, a savings estimate for the model pairs we measured, and a ready request for a free pilot on your own tasks.

Read-only. No key, no account, no telemetry. The server only sends GET requests to agenttrustapi.com, with the User-Agent trust-engine-mcp/<version> (mcp|cli|action) so we can count use per channel; links to agenttrustapi.com in answers carry ?ref=mcp, ?ref=cli or ?ref=action for the same reason. Nothing else is recorded or sent.

Honesty notes. The retirement data is copied from vendor pages, so follow source_url before you act on a date. The benchmark is one coding agent on 27 public exercises: it compares the models fairly, but it does not predict results on your tasks. Savings estimates exist only for pairs we measured; for any other pair the tool gives no number. Nothing is sent for you: request_pilot drafts an e-mail that a person sends.

Connect with one command

Needs Python 3.10+ and uv.

# Claude Codeclaude mcp add --transport stdio trust-engine -- uvx trust-engine-mcp
# Codex CLIcodex mcp add trust-engine -- uvx trust-engine-mcp

Claude Desktop, Cursor and other stdio clients use this config (claude_desktop_config.json, .cursor/mcp.json):

{  "mcpServers": {    "trust-engine": { "command": "uvx", "args": ["trust-engine-mcp"] }  }}

Claude Desktop can also install the one-file bundle trust-engine-mcp-<version>.mcpb from the GitHub releases. Without uv, run pip install trust-engine-mcp and use the command trust-engine-mcp.

Try this first

Are we using any deprecated AI models in this repo?

Use the trust-engine tools. We use claude-sonnet-4-5, gpt-4o and gemini-2.5-flash. Which of them are being retired, on which platform, and by when must we re-test? Cite the sources.

Here is our usage export (CSV). Why are our agents so expensive? List the quick fixes we can do ourselves this week.

Is Claude Sonnet good enough instead of Opus for coding work? Show the benchmark numbers and what we would save on a $4,000 monthly bill. Then tell me how to test this on our own tasks.

Tools

What the scan does not see: model ids built at runtime (from environment variables or string concatenation), models that are not on the calendar, and ids written in a form the calendar does not list. Aliases such as gpt-4.1 match the dated snapshot on the calendar (gpt-4.1-2025-04-14); the answer says so, because an alias can point to a newer snapshot.

Check a repository without an agent

Command line

uvx trust-engine-mcp check            # scan the current folderuvx trust-engine-mcp check src --days 90

It prints each model found, its dates, replacement and source, and where it is used. Exit code 1 if a model found retires within --days days (default 60; a date that has already passed counts), otherwise 0. trust-engine-mcp with no arguments still starts the MCP server.

GitHub Action

# .github/workflows/model-retirements.ymlname: model retirementson:  pull_request:  schedule:    - cron: "17 6 * * 1"   # Mondayspermissions:  contents: readjobs:  check:    runs-on: ubuntu-latest    steps:      - uses: actions/checkout@v4      - uses: DinevDecor/trust-engine-mcp/check@v1        with:          days: "60"                  # default 60          # path: "."                 # default: the whole repository          # fail-on-retiring: "true"  # default false: warnings only

It adds a warning on each line that uses a model retiring within days and a table to the job summary. It needs no token and posts no comments. By default the step never fails; set fail-on-retiring: "true" to fail it.

Usage exports analyze_usage reads

Column names are matched loosely; case, spaces and separators are ignored. Only a model column is required.

• model: model, model_name, model_id, snapshot_id, line_item (e.g. "gpt-4o-2024-08-06, input")
• cost: cost, cost_usd, amount, amount_value, spend, total_cost
• requests: requests, num_model_requests, n_requests, request_count
• tokens: input_tokens, prompt_tokens, n_context_tokens_total, uncached_input_tokens, output_tokens, completion_tokens, cache_read_input_tokens, input_cached_tokens, cache_creation_input_tokens
• reliability: errors, error_count, status, status_code, retries
• other: batch, service_tier, date, start_time, bucket_start

Data sources

The server caches each file for 6 hours with a 10-second timeout. The package includes a snapshot of both files. If the site is unreachable, the server uses that snapshot and says so in every answer.

Environment

Development

pip install -e ".[dev]"pytest && ruff check .

The tests use fixtures and a fake site. They make no network calls.

License

Code MIT © 2026 Trust Engine AI. Retirement data CC BY 4.0. Contact: [email protected]

mcp-name: io.github.DinevDecor/trust-engine-mcp