docs: 10/10 relaunch — zero-dependency story, measurement-type matrix, new tools (Workstream D)

Brings the docs site in line with the v5.0.0 zero-LLM relaunch and documents
every measurement type + the new MCP tools.

- Removed the AI section (docs/ai/*) and its nav; deleted the chat/providers/
  reports/status/overview pages.
- Reframed index.md + mcp/overview.md to the zero-dependency story: no API key,
  no subscription, the MCP agent is the LLM. Tool counts 35 -> 41.
- New docs/reference/measurement-types.md: a matrix of every scan type and
  analysis, its input format, what it computes, and the MCP tool that drives it.
- Documented the 6 new RF-analysis tools in mcp/tools-reference.md and added an
  agent-authored-narrative recipe to mcp/recipes.md.
- reference/api.md: replaced the llm_provider/api_keys sections with the
  analysis_engine + propagation/MIMO function reference.
- glossary.md: dropped the API-key-encryption entries, added diversity/fade/
  Rician terms.
- troubleshooting.md: replaced "AI Summary requires API key" with the
  deterministic + agent-narrative guidance.
- Renamed AI_STATUS.md -> MCP_STATUS.md (rewritten as a zero-LLM MCP tool
  inventory); updated README + the docs.yml path trigger.
- mkdocs build --strict passes (no broken links/nav).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: RFingAdam

.github/workflows/docs.yml

@@ -7,7 +7,7 @@ on:
     paths:
       - 'docs/**'
       - 'mkdocs.yml'
-      - 'AI_STATUS.md'
+      - 'MCP_STATUS.md'
       - 'CONTRIBUTING.md'
       - '.github/workflows/docs.yml'
   workflow_dispatch:

AI_STATUS.md

@@ -0,0 +1,55 @@
+# RFlect MCP — Status & Tool Inventory
+
+**Last Updated**: May 28, 2026
+**Current Version**: v5.0.0
+**Status**: Stable
+
+---
+
+## Zero-dependency by design
+
+As of **v5.0.0**, RFlect makes **no outbound LLM/API calls and requires no API key or paid subscription.** It is a deterministic RF analysis + rendering toolkit. The in-app AI chat assistant, AI report generation, LLM provider abstraction, and encrypted API-key store were all removed.
+
+When RFlect is driven over MCP, the AI agent (Claude Code, Cline, Continue, …) *is* the LLM. RFlect supplies computed data and rendering; the agent supplies any natural-language narrative. Report prose is data-driven by default, or authored by the agent and passed to `generate_report` via the `narrative` parameter.
+
+## MCP server (41 tools, 9 categories)
+
+`rflect-mcp/server.py` exposes the following tool groups. See `docs/mcp/tools-reference.md` for signatures and return shapes.
+
+| Category | Count | Examples |
+|----------|------:|----------|
+| Import | 6 | `import_antenna_file`, `import_passive_pair`, `import_antenna_folder` |
+| Analysis | 5 | `analyze_pattern`, `get_gain_statistics`, `compare_polarizations`, `get_horizon_statistics` |
+| Reports | 3 | `generate_report`, `preview_report`, `get_report_options` |
+| Bulk | 5 | `bulk_process_passive`, `bulk_process_active`, `convert_to_cst` |
+| UWB | 3 | `calculate_sff_from_files`, `analyze_uwb_channel`, `get_impedance_bandwidth` |
+| Calibration Drift | 8 | `cal_drift_ingest`, `cal_drift_compare`, `cal_drift_report` |
+| Orchestration | 1 | `process_folder` *(v4.2.0)* |
+| Validation | 1 | `analyze_iperf_angle_sweep` *(v4.3.0)* |
+| RF Analysis | 6 | `compare_antennas`, `analyze_s11`, `analyze_group_delay`, `estimate_link_budget`, `analyze_mimo_diversity`, `generate_active_cal` *(v5.0.0)* |
+| Misc | 3 | `get_measurement_details`, `validate_file_pair`, help resource |
+
+## Report narrative
+
+`generate_report` produces deterministic, data-driven prose from the measurement
+data with no LLM involved. A driving agent may override any of the following via
+the `narrative` parameter (omitted keys fall back to the deterministic text):
+
+- `executive_summary` (str)
+- `section_analysis` ({measurement_name: str})
+- `recommendations` (str)
+- `captions` ({plot_filename: str})
+
+## Deterministic analysis core
+
+All analysis math lives in `plot_antenna/` (`analysis_engine.py`, `calculations.py`,
+`uwb_analysis.py`, `cal_drift.py`, `file_utils.py`) as pure functions. The MCP tools
+are thin wrappers that validate input, call these functions, and return structured
+dicts. They never make network or LLM calls and never raise — failures surface in a
+`warnings` list.
+
+## History
+
+- **v5.0.0** — Removed the entire AI/LLM stack (chat, AI report generation, provider abstraction, API-key store). Added 6 RF-analysis MCP tools (comparison, S11, group delay, link budget, MIMO diversity, active-cal). `generate_report` gained the agent-authored `narrative` parameter. MCP tool count 35 → 41.
+- **v4.3.0** — `analyze_iperf_angle_sweep` (multi-angle throughput validation).
+- **v4.2.0** — `process_folder` single-call folder orchestration.

MCP_STATUS.md

@@ -127,24 +127,18 @@ python run_rflect.py
 
 - **Polarization Analysis** — Axial ratio, tilt angle, XPD, and polarization sense (LHCP/RHCP) from HPOL/VPOL data with interactive and batch export modes
 - **Batch Processing** — Process an entire folder of HPOL/VPOL pairs or TRP files automatically, with organized per-pair output
-- **Report Generation** — Export DOCX reports with embedded plots, measurement summaries, and optional AI-generated executive analysis
+- **Report Generation** — Export DOCX reports with embedded plots, measurement summaries, and deterministic data-driven prose (or narrative authored by the driving MCP agent)
 - **3D Visualization** — Perceptually uniform turbo colormap, transparent panes, coordinate axes, and manual or auto Z-axis scaling
 
-## AI-Powered Analysis (Optional)
+## Zero-dependency, MCP-driven (v5.0.0)
 
-RFlect integrates with **OpenAI**, **Anthropic**, and **Ollama** to bring intelligent analysis to your measurement workflow. Ask questions about your data in natural language, get AI-generated executive summaries in reports, or let vision models interpret your radiation patterns.
+RFlect makes **no outbound LLM/API calls and needs no API key or subscription.** It is a deterministic RF analysis + rendering toolkit. When driven over MCP, the AI agent *is* the LLM: it calls RFlect's tools for data and — if a report needs narrative prose — authors it itself and passes it to `generate_report`. Everything RFlect computes is reproducible, not generated.
 
-All AI features are completely optional — RFlect works fully without any provider configured. When you're ready, configure a provider via **Tools > Manage API Keys**. Your keys are encrypted locally with AES-128 and never leave your machine.
-
-<p align="center">
-  <img src="./assets/api_key_management.png" alt="API Key Management" width="440">
-</p>
-
-See [AI_STATUS.md](AI_STATUS.md) for provider details and supported models.
+See [MCP_STATUS.md](MCP_STATUS.md) for the full tool inventory.
 
 ## MCP Server
 
-RFlect ships with an [MCP](https://modelcontextprotocol.io/) server — 25 tools that let AI assistants like Claude Code import your measurements, run analysis, generate reports, and perform UWB characterization programmatically. No GUI required.
+RFlect ships with an [MCP](https://modelcontextprotocol.io/) server — 41 tools that let an AI agent like Claude Code import your measurements, run analysis, compare antennas, estimate link budgets, generate reports, and perform UWB/MIMO characterization programmatically. No GUI required.
 
 See [rflect-mcp/README.md](rflect-mcp/README.md) for setup and the full tool reference.