Submit
lluvr

Frame Check

Community lluvr
Updated

Structural framing analysis as an MCP server with deterministic detection. The caller's agent runs the judge, so Frame Check bears zero LLM cost per query.

Frame Check

See what any document does not show you.

Frame Check is a structural framing analysis tool. It names whichperspectives a document takes, which it omits, and how it positionsthe reader. Numerical claims are cross-checked against authoritativesources where coverage exists.

Quickstart (MCP server)

The PyPI package frame-check-mcp is the Model Context Protocolserver. It runs locally and gives any MCP-compatible AI client(Claude Desktop, Cursor, Cline, Continue.dev, etc.) deterministicstructural framing analysis as a tool.

pip install frame-check-mcp

Then point your MCP client at the installed entry point. ForClaude Desktop, add to claude_desktop_config.json:

{
  "mcpServers": {
    "frame-check": {
      "command": "frame-check-mcp"
    }
  }
}

Restart the client. Then in any conversation: "Can you frame-checkthis document?" Full install + verification details in MCP_SERVER.md.

What it does

Pass a document and Frame Check returns:

  • A structural framing profile: which of five analytical perspectives(causes, risks, stakeholders, trends, uncertainty) the documentcovers, which it omits, and the density of each.
  • Voice and epistemic posture: how the document positions the reader,and what share of claims are attributed to sources.
  • Temporal orientation: whether the document grounds its conclusionsin historical data, present state, or projections.
  • Frame Vocabulary Standard candidate matches: named frame patternswhose rule-based signals fire on the text, each with identificationcues and worked examples. Matches are candidate-level; precisionagainst multi-source labeling is an active research question.
  • Source-network verification: numeric claims checked against SECEDGAR, FRED, World Bank, REST Countries, Alpha Vantage, and WolframAlpha where those providers have coverage.
  • An optional AI narrative interpreting framing at prose level.Labelled distinctly so readers do not conflate language-modelinterpretation with deterministic measurement.

Approach

Structural measurement is the floor. Every framing claim the toolmakes is computed from deterministic pattern matchers and alwaysreturns the same result for the same input. AI-assisted interpretationis available as enrichment where an API key is configured, but islabelled as such and never hidden behind the structural layer.

Verification is bounded. The tool only verifies numeric claims againstproviders with genuine coverage for the claim type, and it surfacesits own calibration results (precision, recall, F1 per provider)rather than asserting verdicts without evidence.

Named-pattern detection is a separate reliability layer from thestructural profile. Detector F1 = 0.36 against expert labelers in apre-registered validation, below the useful threshold of 0.4. Thepivot to construct-honesty surfacing (under-detection markers,density caveats, confidence states) rather than confident labels isthe load-bearing claim at v0; a reader-aid study (Track B,pre-registered) tests whether this surfacing actually helps a readersee framing they would otherwise miss.

Honest limits and anticipated adversarial readings are catalogued inANTICIPATED_CRITIQUES.md.

Documentation

  • METHODOLOGY.md: full methodology paper (v0.2 draft)
  • MCP_SERVER.md: MCP server reference (tools, resources, prompts)
  • MCP_CLIENT_CONFORMANCE_v1.md: 32/32 conformance round-trips against the installed wheel
  • FRAME_DIVERGENCE_v1.md and FRAME_DIVERGENCE_v2.md: divergence architecture
  • data/frame_library/: 20-entry FVS catalog with worked examples
  • ANTICIPATED_CRITIQUES.md: self-enumerated adversarial readings
  • VALIDATION_PROGRAM.md: observational + formal validation plans

Running tests

pip install -r requirements.txt
python3 run_tests.py

49 test suites, ~3 minutes. Includes 63 adversarial dispatcher testsin test_mcp_adversarial.py plus the V4.2 engine + classifier coverage.

License

Apache-2.0 for code; CC-BY-4.0 for the FVS library and worked examples(see NOTICE for the per-directory enumeration).

Citation

If Frame Check is useful in your work, see CITATION.cff for thecitable form. Frame Check is authored by Lovro Lucic; named authorshipis the project's primary credibility asset per the construct-honestydiscipline.

Contributing

Sign-off-by-DCO required per CONTRIBUTING.md. Governance perGOVERNANCE.md (BDFL model with named forcing functions forcanon-promotion decisions). External reviewer engagement perREVIEWERS.md.

Issues

https://github.com/lluvr/frame-check-mcp/issues

MCP Server · Populars

MCP Server · New