{"id":27,"date":"2026-09-22T04:49:55","date_gmt":"2026-09-22T04:49:55","guid":{"rendered":"https:\/\/spectrcyde.com\/?p=27"},"modified":"2026-09-22T04:49:55","modified_gmt":"2026-09-22T04:49:55","slug":"from-visualization-to-scientific-instrument-scythes-contextual-mcp-workbench","status":"publish","type":"post","link":"https:\/\/spectrcyde.com\/?p=27","title":{"rendered":"From Visualization to Scientific Instrument: SCYTHE’s Contextual MCP Workbench"},"content":{"rendered":"\n

Date:<\/strong> August 14, 2026 \u00a0<\/p>\n\n\n\n

Author:<\/strong> SCYTHE Team \u00a0<\/p>\n\n\n\n

Audience:<\/strong> National Laboratory Research Institutions \u00a0<\/p>\n\n\n\n

Category:<\/strong> GraphOps, Scientific Computing, Human\u2013Machine Teaming, Evidence-Centered Operations<\/p>\n\n\n\n

—<\/strong><\/p>\n\n\n\n

Scientific visualization becomes substantially more valuable when an operator<\/p>\n\n\n\n

can move from a visible anomaly to the exact computational capability needed to<\/p>\n\n\n\n

investigate it\u2014without losing provenance, widening the evidence scope, or<\/p>\n\n\n\n

silently authorizing an action.<\/p>\n\n\n\n

That is the purpose of SCYTHE’s latest Clarktech advancement: a contextual<\/p>\n\n\n\n

Model Context Protocol workbench embedded directly within the live RF and<\/p>\n\n\n\n

network hypergraph instrument.<\/p>\n\n\n\n

The workbench connects SCYTHE-Web to the considerable machinery already<\/p>\n\n\n\n

running inside a SCYTHE server: live graph analytics, GraphOps Copilot,<\/p>\n\n\n\n

autonomous observation patrols, RF instrumentation, semantic memory, tactical<\/p>\n\n\n\n

event history, and guarded control capabilities. It does so through a narrow,<\/p>\n\n\n\n

server-owned interface rather than exposing a general-purpose tool console to<\/p>\n\n\n\n

the browser.<\/p>\n\n\n\n

The result is a transition from dashboard to scientific instrument.<\/p>\n\n\n\n

“`text<\/p>\n\n\n\n

Visible phenomenon<\/p>\n\n\n\n

        |<\/p>\n\n\n\n

        v<\/p>\n\n\n\n

Typed, revision-bearing selection<\/p>\n\n\n\n

        |<\/p>\n\n\n\n

        v<\/p>\n\n\n\n

Contextual workbench panel<\/p>\n\n\n\n

        |<\/p>\n\n\n\n

        v<\/p>\n\n\n\n

Fixed read-only MCP projection<\/p>\n\n\n\n

        |<\/p>\n\n\n\n

        v<\/p>\n\n\n\n

Bounded evidence + tool receipt<\/p>\n\n\n\n

        |<\/p>\n\n\n\n

        v<\/p>\n\n\n\n

Revision-retaining GraphOps investigation<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

Open the instrument at:<\/p>\n\n\n\n

“`text<\/p>\n\n\n\n

http:\/\/127.0.0.1:5001\/scythe-web\/regional-rf-demo.html<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

## Why This Matters to Research Institutions<\/strong><\/p>\n\n\n\n

National laboratories routinely work across boundaries that conventional<\/p>\n\n\n\n

analytics products flatten together:<\/p>\n\n\n\n

– measurement versus model output;<\/p>\n\n\n\n

– data-plane behavior versus control-plane observation;<\/p>\n\n\n\n

– self-reported infrastructure versus independently observed infrastructure;<\/p>\n\n\n\n

– live state versus a retained historical revision;<\/p>\n\n\n\n

– interpretation versus execution authority;<\/p>\n\n\n\n

– locally held evidence versus deliberately disclosed external evidence.<\/p>\n\n\n\n

SCYTHE treats these distinctions as part of the data model and interaction<\/p>\n\n\n\n

contract. A model can interpret evidence, but it cannot promote an inference<\/p>\n\n\n\n

into an observation. A visualization can focus attention, but it cannot alter<\/p>\n\n\n\n

the authoritative value. An MCP tool can describe an available action, but a<\/p>\n\n\n\n

browser panel cannot bypass the orchestrator’s proposal and approval boundary.<\/p>\n\n\n\n

This is particularly important in research environments where a compelling<\/p>\n\n\n\n

visual explanation is not sufficient. A result must remain inspectable,<\/p>\n\n\n\n

repeatable, and falsifiable.<\/p>\n\n\n\n

## Forty-Eight Capabilities, Four Bounded Windows<\/strong><\/p>\n\n\n\n

The active SCYTHE server currently exposes 48 MCP tools:<\/p>\n\n\n\n

| Capability family | Tool count | Representative functions |<\/p>\n\n\n\n

|—|—:|—|<\/p>\n\n\n\n

| Core graph operations | 24 | hot entities, recent edges, scope statistics, temporal scrubbing, graph export |<\/p>\n\n\n\n

| GraphOps Copilot | 7 | bounded investigation, directive compilation, entity parsing, coverage explanation |<\/p>\n\n\n\n

| GraphOps Autopilot | 5 | patrol status, suggestions, observations, analyst cards, feedback |<\/p>\n\n\n\n

| RF evidence and control | 7 | bridge status, spectrum summaries, RF queries, correlation, guarded tuning and capture |<\/p>\n\n\n\n

| Semantic memory | 5 | entity embedding, similarity search, anomaly matching, identity stitching, clustering |<\/p>\n\n\n\n

SCYTHE-Web does not present those tools as a flat command catalog. The new<\/p>\n\n\n\n

workbench organizes the read-only portion into four operator-facing panels:<\/p>\n\n\n\n

### AUTOPILOT<\/strong><\/p>\n\n\n\n

The Autopilot panel exposes the GraphOps Sentinel’s runtime state, queued<\/p>\n\n\n\n

suggestions, and low-confidence observation log.<\/p>\n\n\n\n

This makes autonomous patrol visible without granting it autonomous authority.<\/p>\n\n\n\n

An operator can inspect why a pattern was raised, select implicated entities,<\/p>\n\n\n\n

and carry the finding into a GraphOps investigation. Analyst feedback remains<\/p>\n\n\n\n

a guarded capability because it can affect downstream training and TAK-ML<\/p>\n\n\n\n

workflows.<\/p>\n\n\n\n

“`text<\/p>\n\n\n\n

AUTOPILOT \/\/ OK \/\/ 3\/3 READ TOOLS<\/p>\n\n\n\n

MCP \/\/ graphops_autopilot_status<\/p>\n\n\n\n

MCP \/\/ graphops_suggestion_queue<\/p>\n\n\n\n

MCP \/\/ graphops_observation_log<\/p>\n\n\n\n

GUARDED CAPABILITY<\/p>\n\n\n\n

graphops_submit_feedback \/\/ ANALYST REVIEW REQUIRED<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

### SEMANTIC<\/strong><\/p>\n\n\n\n

The Semantic panel provides bounded access to SCYTHE’s FAISS-backed semantic<\/p>\n\n\n\n

memory. When a graph entity is selected and the semantic corpus contains<\/p>\n\n\n\n

vectors, the panel can retrieve behaviorally similar entities while retaining<\/p>\n\n\n\n

the selected graph revision.<\/p>\n\n\n\n

The current system also handles the scientifically honest empty state. If<\/p>\n\n\n\n

semantic memory contains zero vectors, it reports that fact and does not<\/p>\n\n\n\n

repeatedly invoke an embedding model to manufacture an appearance of insight.<\/p>\n\n\n\n

Admission of a selected entity into semantic memory is shown as a proposal<\/p>\n\n\n\n

requiring an explicit corpus decision.<\/p>\n\n\n\n

This creates a foundation for questions such as:<\/p>\n\n\n\n

– Have we observed a structurally similar route anomaly before?<\/p>\n\n\n\n

– Which entities share a behavioral signature across collection sessions?<\/p>\n\n\n\n

– Is the selected event near an established cluster or genuinely novel?<\/p>\n\n\n\n

– What evidence distinguishes the nearest analogue from the present case?<\/p>\n\n\n\n

Similarity remains a retrieval aid. It is not identity, attribution, or<\/p>\n\n\n\n

causality.<\/p>\n\n\n\n

### SPECTRUM<\/strong><\/p>\n\n\n\n

The Spectrum panel exposes the RF bridge state, the latest bounded FFT summary,<\/p>\n\n\n\n

and recent RF observations. Raw IQ is not returned through this interface.<\/p>\n\n\n\n

“`text<\/p>\n\n\n\n

MCP \/\/ rf_bridge_status<\/p>\n\n\n\n

MCP \/\/ rf_spectrum_snapshot \/\/ BOUNDED FFT SUMMARY<\/p>\n\n\n\n

MCP \/\/ rf_observations_query \/\/ OBSERVED<\/p>\n\n\n\n

GUARDED CAPABILITIES<\/p>\n\n\n\n

rf_tune            \/\/ ORCHESTRATOR APPROVAL REQUIRED<\/p>\n\n\n\n

rf_capture_control \/\/ ORCHESTRATOR APPROVAL REQUIRED<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

This distinction supports laboratory workflows in which observation and<\/p>\n\n\n\n

instrument control may have different authorization, safety, scheduling, and<\/p>\n\n\n\n

configuration-management requirements. The display can show that tuning or<\/p>\n\n\n\n

capture is possible without making the visualization itself a control surface.<\/p>\n\n\n\n

### EVENTS<\/strong><\/p>\n\n\n\n

The Events panel provides a compact operational view of graph metrics, highly<\/p>\n\n\n\n

connected entities, recent edges, and the bounded neighborhood of the selected<\/p>\n\n\n\n

entity.<\/p>\n\n\n\n

Entity references returned by MCP become selectable objects. Selecting one<\/p>\n\n\n\n

focuses the same entity across the live hypergraph, infrastructure lens, graph<\/p>\n\n\n\n

explorer, and GraphOps dialog. The system therefore connects multiple<\/p>\n\n\n\n

representations without creating separate, conflicting selection states.<\/p>\n\n\n\n

Event ingestion and TAK-ML execution remain guarded proposals.<\/p>\n\n\n\n

## A Deliberately Narrow MCP Boundary<\/strong><\/p>\n\n\n\n

The browser does not receive a generic `tools\/call` facility. Instead, it sends<\/p>\n\n\n\n

one of four panel names plus an optional typed graph selection to the<\/p>\n\n\n\n

orchestrator:<\/p>\n\n\n\n

“`json<\/p>\n\n\n\n

{<\/p>\n\n\n\n

  “panel”: “events”,<\/p>\n\n\n\n

  “selection”: {<\/p>\n\n\n\n

    “kind”: “graph-node”,<\/p>\n\n\n\n

    “entityId”: “host:203.0.113.8”,<\/p>\n\n\n\n

    “graphRevision”: “graph-a”<\/p>\n\n\n\n

  }<\/p>\n\n\n\n

}<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

The server\u2014not the browser\u2014chooses the exact MCP tools and arguments associated<\/p>\n\n\n\n

with that panel. Unknown request fields and unsupported panels are refused.<\/p>\n\n\n\n

Returned dictionaries, arrays, strings, and nesting depth are bounded before<\/p>\n\n\n\n

they reach the visualization.<\/p>\n\n\n\n

This architecture provides several useful properties:<\/p>\n\n\n\n

1. **Capability confinement.**<\/strong> Adding a tool to the MCP registry does not<\/p>\n\n\n\n

   automatically make it browser-callable.<\/p>\n\n\n\n

2. **Argument ownership.**<\/strong> The server controls time windows, collection limits,<\/p>\n\n\n\n

   similarity counts, and whether FFT bins are included.<\/p>\n\n\n\n

3. **Mutation isolation.**<\/strong> Direct execution of mutating MCP tools remains<\/p>\n\n\n\n

   disabled; guarded capabilities are descriptive proposals only.<\/p>\n\n\n\n

4. **Reduced UI authority.**<\/strong> Browser code cannot silently transform a<\/p>\n\n\n\n

   read-oriented scientific display into an administration console.<\/p>\n\n\n\n

5. **Auditable evolution.**<\/strong> Each new exposed capability requires an explicit<\/p>\n\n\n\n

   server-side mapping and test.<\/p>\n\n\n\n

This is defense in depth, not a claim of formal security certification. A<\/p>\n\n\n\n

production deployment still requires institutional authentication,<\/p>\n\n\n\n

authorization, network segmentation, audit retention, software assurance, and<\/p>\n\n\n\n

configuration control appropriate to its environment.<\/p>\n\n\n\n

## Revision Retention Without Temporal Overclaiming<\/strong><\/p>\n\n\n\n

Graph selections carry the revision that the operator actually saw. When a<\/p>\n\n\n\n

workbench result is opened in GraphOps, that selection revision is retained in<\/p>\n\n\n\n

the investigation tab.<\/p>\n\n\n\n

MCP results, however, are live observations unless a particular tool returns a<\/p>\n\n\n\n

historical snapshot. SCYTHE states this difference directly:<\/p>\n\n\n\n

“`text<\/p>\n\n\n\n

BOUNDARY \/\/ LIVE OBSERVATIONAL MCP RESULTS;<\/p>\n\n\n\n

THE SELECTION REVISION IS RETAINED,<\/p>\n\n\n\n

BUT RESULTS ARE NOT HISTORICAL SNAPSHOT CLAIMS<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

This small distinction prevents a common analytical error: displaying a<\/p>\n\n\n\n

historical selection beside a current metric and implying that both describe<\/p>\n\n\n\n

the same instant.<\/p>\n\n\n\n

For laboratory reproducibility, the next logical refinement is a receipt that<\/p>\n\n\n\n

records the selected graph revision, MCP tool version, normalized arguments,<\/p>\n\n\n\n

result hash, server generation time, and any retained dataset identifiers. The<\/p>\n\n\n\n

current workbench establishes the interaction boundary needed for that richer<\/p>\n\n\n\n

receipt.<\/p>\n\n\n\n

## One Selection, Several Scientific Representations<\/strong><\/p>\n\n\n\n

The workbench participates in SCYTHE’s shared selection fabric.<\/p>\n\n\n\n

“`text<\/p>\n\n\n\n

                     +–> 3D causal chamber<\/p>\n\n\n\n

                     |<\/p>\n\n\n\n

Selected entity —–+–> 2D accessible topology<\/p>\n\n\n\n

                     |<\/p>\n\n\n\n

                     +–> Cesium location estimate<\/p>\n\n\n\n

                     |<\/p>\n\n\n\n

                     +–> InfraFlow evidence layers<\/p>\n\n\n\n

                     |<\/p>\n\n\n\n

                     +–> Graph Explorer neighborhood<\/p>\n\n\n\n

                     |<\/p>\n\n\n\n

                     +–> Contextual MCP workbench<\/p>\n\n\n\n

                     |<\/p>\n\n\n\n

                     +–> Revision-bearing GraphOps tab<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

An Autopilot finding can therefore become a highlighted graph entity. A hot<\/p>\n\n\n\n

entity returned by the Events panel can become a bounded host investigation.<\/p>\n\n\n\n

A semantic analogue can be selected and compared without copying identifiers<\/p>\n\n\n\n

between interfaces. A Spectrum observation can be interpreted alongside an RF<\/p>\n\n\n\n

cell while its evidence class remains visible.<\/p>\n\n\n\n

The visual effect is useful, but the more consequential outcome is analytical<\/p>\n\n\n\n

continuity: each representation refers to the same typed object rather than an<\/p>\n\n\n\n

informal label reconstructed by the user.<\/p>\n\n\n\n

## Building on the InfraFlow and Full-Fidelity Foundation<\/strong><\/p>\n\n\n\n

This release builds on SCYTHE’s recent infrastructure and reasoning work:<\/p>\n\n\n\n

– normalized Suricata summaries enter a bounded live hypergraph;<\/p>\n\n\n\n

– adaptive relevance separates detected graph size from displayed graph size;<\/p>\n\n\n\n

– host liveness remains independent of analytical purpose;<\/p>\n\n\n\n

– traceroute measurements remain distinct from inferred GeoIP geography;<\/p>\n\n\n\n

– PeeringDB declarations and RIPE RIS control-plane observations remain<\/p>\n\n\n\n

  parallel to the measured data plane;<\/p>\n\n\n\n

– contradictions are preserved rather than averaged away;<\/p>\n\n\n\n

– Full-Fidelity Evidence Capsules can disclose exact bounded evidence to an<\/p>\n\n\n\n

  explicitly selected cloud model while excluding credentials, raw packet<\/p>\n\n\n\n

  payloads, unrelated files, and directive authority;<\/p>\n\n\n\n

– local Ollama remains available for private, interpretive investigations.<\/p>\n\n\n\n

The contextual workbench gives those layers a common operational doorway. It<\/p>\n\n\n\n

does not collapse them into a single confidence score or a single AI answer.<\/p>\n\n\n\n

## Verification<\/strong><\/p>\n\n\n\n

The implementation was tested at the server, module, browser, and live-service<\/p>\n\n\n\n

levels:<\/p>\n\n\n\n

“`text<\/p>\n\n\n\n

GraphOps and server tests           56 passed<\/p>\n\n\n\n

SCYTHE-Web unit tests               85 passed<\/p>\n\n\n\n

Packaged Chromium integration tests 3 passed<\/p>\n\n\n\n

Live Chromium acceptance            passed with no console errors<\/p>\n\n\n\n

Live workbench panels                AUTOPILOT \/ SEMANTIC \/ SPECTRUM \/ EVENTS<\/p>\n\n\n\n

Mutation path                        unavailable through workbench endpoint<\/p>\n\n\n\n

“`<\/p>\n\n\n\n

The browser integration test verifies more than rendering. It confirms that:<\/p>\n\n\n\n

– the selected entity and graph revision enter the workbench request;<\/p>\n\n\n\n

– bounded MCP evidence appears in the panel;<\/p>\n\n\n\n

– guarded capabilities are visibly separated;<\/p>\n\n\n\n

– `OPEN IN GRAPHOPS` transfers the evidence into an investigation;<\/p>\n\n\n\n

– the original graph revision survives that transition.<\/p>\n\n\n\n

## Research Directions<\/strong><\/p>\n\n\n\n

For national laboratory environments, this architecture opens several<\/p>\n\n\n\n

practical research directions.<\/p>\n\n\n\n

### Reproducible Instrument Receipts<\/strong><\/p>\n\n\n\n

Bind every workbench projection to content hashes, tool versions, model<\/p>\n\n\n\n

identities, clock sources, and retained graph revisions. This would allow an<\/p>\n\n\n\n

investigation to be replayed or compared across software baselines.<\/p>\n\n\n\n

### Facility-Local Capability Profiles<\/strong><\/p>\n\n\n\n

Define institution-controlled profiles for air-gapped, restricted-network,<\/p>\n\n\n\n

and externally connected deployments. The same user interface could expose a<\/p>\n\n\n\n

different allow-list without changing evidence semantics.<\/p>\n\n\n\n

### Multi-Instrument Correlation<\/strong><\/p>\n\n\n\n

Extend the selection fabric across RF receivers, network sensors, AIS,<\/p>\n\n\n\n

geospatial layers, optical instruments, and simulation products. Compatibility<\/p>\n\n\n\n

checks should refuse cross-domain correlation when time, location, calibration,<\/p>\n\n\n\n

or provenance requirements do not align.<\/p>\n\n\n\n

### Human\u2013Machine Teaming Studies<\/strong><\/p>\n\n\n\n

Measure whether contextual capability presentation reduces operator error,<\/p>\n\n\n\n

time-to-falsifier, and unsupported causal conclusions compared with a flat tool<\/p>\n\n\n\n

catalog or unconstrained conversational interface.<\/p>\n\n\n\n

### HPC and Campaign-Scale Analysis<\/strong><\/p>\n\n\n\n

Keep interactive selection and provenance local while routing approved,<\/p>\n\n\n\n

content-addressed analytical jobs to institutional clusters. Results should<\/p>\n\n\n\n

return as evidence-bearing artifacts, not detached prose.<\/p>\n\n\n\n

### Formal Policy Enforcement<\/strong><\/p>\n\n\n\n

Move the current allow-list and proposal boundaries toward declarative policy:<\/p>\n\n\n\n

tool authority, data classification, operator role, instrument state, network<\/p>\n\n\n\n

destination, and required approvals could all become machine-verifiable inputs<\/p>\n\n\n\n

to execution decisions.<\/p>\n\n\n\n

## The Larger Principle<\/strong><\/p>\n\n\n\n

The central advance is not that SCYTHE can call 48 tools.<\/p>\n\n\n\n

It is that a researcher can select a phenomenon, see only the capabilities that<\/p>\n\n\n\n

make sense for that context, inspect the evidence each capability returns, and<\/p>\n\n\n\n

carry the result into a continuing investigation\u2014while the system preserves<\/p>\n\n\n\n

the boundaries between observation, inference, interpretation, and action.<\/p>\n\n\n\n

For research institutions, that boundary is not friction. It is part of the<\/p>\n\n\n\n

instrument.<\/p>\n\n\n\n

SCYTHE’s visual language now reaches the server’s deeper machinery. The<\/p>\n\n\n\n

machine has become more capable, but its claims have not become less<\/p>\n\n\n\n

accountable.<\/p>\n","protected":false},"excerpt":{"rendered":"

Date: August 14, 2026 \u00a0 Author: SCYTHE Team \u00a0 Audience: National Laboratory Research Institutions \u00a0 Category: GraphOps, Scientific Computing, Human\u2013Machine Teaming, Evidence-Centered Operations — Scientific visualization becomes substantially more valuable when an operator can move from a visible anomaly to the exact computational capability needed to investigate it\u2014without losing provenance, widening the evidence scope, or […]<\/p>\n","protected":false},"author":1,"featured_media":28,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-27","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-our-friends"],"_links":{"self":[{"href":"https:\/\/spectrcyde.com\/index.php?rest_route=\/wp\/v2\/posts\/27","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/spectrcyde.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/spectrcyde.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/spectrcyde.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/spectrcyde.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=27"}],"version-history":[{"count":1,"href":"https:\/\/spectrcyde.com\/index.php?rest_route=\/wp\/v2\/posts\/27\/revisions"}],"predecessor-version":[{"id":29,"href":"https:\/\/spectrcyde.com\/index.php?rest_route=\/wp\/v2\/posts\/27\/revisions\/29"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/spectrcyde.com\/index.php?rest_route=\/wp\/v2\/media\/28"}],"wp:attachment":[{"href":"https:\/\/spectrcyde.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=27"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/spectrcyde.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=27"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/spectrcyde.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=27"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}