Tools and Compatibility

origin-mcp exposes MCP tools for Origin project management, worksheet editing, plotting, graph formatting, analysis, export, and lifecycle control. By default it uses a compact tool profile to reduce model tool-selection cost.

Tool failures return ok=false, a human-readable message, the Python error_type, and a stable error_code such as worksheet_not_found, graph_not_found, file_not_found, path_not_allowed, unsupported_origin_feature, unsupported_analysis_type, or origin_dependency_unavailable. They also include a recoverable boolean and an ordered next_actions list with concrete recovery steps. Clients should branch on error_code, present or execute safe next_actions, and retry only when recoverable=true and the failed operation is safe to repeat. Successful responses omit both recovery fields.

{
  "ok": false,
  "message": "Worksheet not found: [Book1]Data",
  "error_code": "worksheet_not_found",
  "recoverable": true,
  "next_actions": [
    "Inspect the current Origin project objects and correct the referenced name or ID.",
    "Create or import the missing object before retrying the operation."
  ],
  "data": {
    "error_type": "OriginOperationError",
    "error_code": "worksheet_not_found"
  }
}

Tool Profiles

The default profile is compact, which registers 25 high-level tools (listed in COMPACT_TOOL_NAMES and reported by origin_doctor). It keeps the common workflow surface small while preserving specialized wrappers as internal Python functions.

Set ORIGIN_MCP_TOOL_PROFILE before starting the MCP server to select a larger surface: data adds worksheet transforms; plot adds specialized plot, graph, template, and multi-panel composition tools; analysis adds typed fitting, result inspection/recalculation, and advanced-statistics tools; and standard restores the previous curated surface. full exposes every wrapper; expert and all are aliases for it. Unknown values safely fall back to compact.

Default Compact Tools

The exact compact tool list is owned by COMPACT_TOOL_NAMES in src/origin_mcp/tools/_shared.py and is reported by origin_doctor. The searchable tool catalog is generated from tool module docstrings and is available through the mcp_tools knowledge collection.

In compact mode, use origin_query_knowledge or origin_browse_knowledge to discover the right high-level workflow instead of choosing from every specialized wrapper.

Named plotting calls are idempotent by default where Origin exposes an output graph layer target. When a table plotting tool is called with graph_name, origin-mcp first looks for that graph page, clears its existing plots, and draws the new result into the same page instead of creating Graph2, Graph3, and so on. If no book_name is supplied, the imported data is stored in a stable workbook named from graph_name plus _Data; repeated calls overwrite that worksheet. Calls without explicit names still create fresh Origin objects. A few worksheet-command Plot Type ID routes that do not expose an output graph target still use Origin's native creation route and do not clear an existing graph.

Parameterized plotting with origin_plot

origin_plot(path, kind, ...) is a single compact-profile entry point for the table plot kinds that each also have a dedicated origin_plot_* wrapper in the full profile. It lets compact-mode clients reach every one of these kinds without enabling ORIGIN_MCP_TOOL_PROFILE=full. kind must be one of: area, stack_area, fill_area, bar, stack_bar, floating_bar, column_stack, pie, ternary, ternary_contour, bubble, bubble_color_mapped, color_mapped, vector_xyam, vector_xyxy, vector_3d, high_low_close, candlestick, waterfall, ribbon_3d, bars_3d, errorbar_3d, polar_xr_ytheta, smith, or dendrogram, plus the common line, scatter, line_symbol, column, histogram, and box kinds. For common XY plots, the first selected column is X and the remainder are Y; a single selected column is treated as Y-only. An unknown kind returns ok=false with error_code=invalid_request and lists the valid kinds. For matrix-range plots use origin_plot_matrix_id from the plot/full profile. The entry point accepts the shared selected_cols, graph_name, title, export_path, and style_mode arguments and follows the same idempotent graph_name behavior described above.

{"path": "data/processed/sales.csv", "kind": "bar", "graph_name": "Sales"}

Visual verification with origin_view_graph

origin_view_graph(graph_name=None, max_width=1600) renders a graph and returns it as an image content block the model can actually see, so a vision-capable client can visually verify a plot and iterate on it. Unlike origin_export_graph and origin_export_preview, it leaves no file behind: the graph is rendered to a temporary PNG, returned inline alongside a short text summary (graph name, pixel dimensions, byte size), and the temp file is deleted. max_width bounds the rendered pixel width to keep the image — and its token cost — small. Pass a graph_name to target a specific page, or omit it to render the active graph. Use origin_export_graph instead when you need a persistent file on disk.

User Template Library

origin-mcp can save a finished graph as a reusable Origin template and find a matching one before plotting, so a user's preferred styling is captured once and reapplied to same-type figures.

Templates live in a per-user library, by default ~/.origin-mcp/templates (override with ORIGIN_MCP_TEMPLATE_DIR). Each saved template is stored as three files plus a shared index: <slug>.otpu (the Origin graph template), <slug>.json (searchable metadata), and <slug>.png (a preview thumbnail), aggregated into index.json.

The save tool stores the active or named graph and accepts optional plot_types (e.g. ["scatter"]), roles (e.g. ["x", "y"]), tags, description, and n_columns that make the template easier to match later; the layer and plot counts are captured automatically. The search tool ranks the library against an intended plot and returns each candidate with a score and human-readable match_reasons. Ranking weights an exact plot-type match highest, then a same plot-type family (e.g. a line_symbol template for a scatter request), close column counts, matching tags, and keyword overlap with the name, description, and tags, returning an empty list when nothing matches. A list tool reports every saved template most recent first, a delete tool removes a template's files and index entry, a rename tool moves a template to a new name (its .otpu/.json/ .png files and index entry) without redrawing, and a metadata-update tool edits a template's description, tags, and matching hints in place without touching the .otpu or needing a live graph. These management tools report not_found when no template carries the given name.

Once saved, reuse a template by passing its name to any plotting tool's template argument (for example a table plot tool's template or a FigureSpec style.template). A bare name is resolved against the library first, then falls back to Origin's built-in templates and explicit file paths.

Searching, listing, and deleting operate purely on the local library files and do not require Origin to be running; only saving a template drives Origin. The template management tools live in the plot, standard, and full profiles. Their exact names are available in the generated mcp_tools knowledge collection; discover them with origin_query_knowledge.

FigureSpec Tools

origin-mcp includes a first-pass declarative FigureSpec workflow for agent-friendly plotting plans. A FigureSpec describes the desired figure in terms of data, page, layer, plot, annotation, style, export, and QA sections. The MCP server then validates the structure and translates the supported subset into existing Origin plotting and formatting calls.

Use origin_plan_figure_spec(spec) to validate a JSON FigureSpec and return the planned Origin operations without touching Origin. Planning reads the data file headers and verifies that mapped columns and column indexes exist before any Origin calls are made. Use origin_execute_figure_spec(spec, dry_run=false) to execute the current supported subset: worksheet-backed single-panel, grid, custom grid/span, inset, and dual-Y figures; common plot types; X/Y/Z axis settings and one X/Y break per axis; panel/legend/reference annotations; exports, OPJU save, and graph diagnostics. Dual-Y specs use Origin's doubleY template and require two layers, one shared data source and X mapping, and one common line/scatter plot type without uncertainty mappings. Unsupported features are reported in the plan instead of being guessed.

Minimal JSON shape:

{
  "figure": {"id": "line_demo", "title": "Line Demo"},
  "data": [
    {
      "id": "ds_line",
      "source": "data/processed/line.csv",
      "object": "worksheet",
      "roles": {"x": "time", "y": "response"}
    }
  ],
  "page": {"layout": "grid"},
  "layers": [
    {
      "id": "panel_a",
      "data_ref": "ds_line",
      "grid_cell": [0, 0],
      "x": {"title": "Time (s)", "limits": "auto"},
      "y": {"title": "Response", "limits": "auto"},
      "panel_tag": "(a)"
    }
  ],
  "plots": [
    {
      "id": "plot_a",
      "layer": "panel_a",
      "type": "line",
      "map": {"x": "time", "y": "response"}
    }
  ],
  "style": {"theme": "nature", "palette_name": "nature"},
  "export": {
    "dir_figures": "output/figures",
    "dir_opju": "output/opju",
    "png": {"enabled": true},
    "pdf": {"enabled": true},
    "qa": {"require_opju": true, "require_axis_titles": true}
  }
}

For custom panel layouts, set page.layout to custom. Grid-style custom layouts use page.size_mm, optional page.margins_mm, optional page.panel_spacing_mm, and per-layer grid_cell / grid_span; the executor converts those values into Origin page size plus per-layer geometry. Absolute custom layouts use position_mode="absolute" plus position.left, position.top, position.width, and position.height as page percentages. For an inset, use page.layout="inset"; absolute positions are honored, while unspecified positions get a deterministic upper-right inset. Use page.layout="dual_y" with the first layer as the left axis and the second as the right axis.

For Nature-style graph formatting, origin_palette_catalog() lists the built-in palette registry, including semantic roles, source links, color counts, and license notes. Each entry also includes accessibility: white-page contrast, OKLab perceptual separation, and minimum separation under protanopia, deuteranopia, and tritanopia simulations. screening_status="pass" means the palette clears the automated preflight thresholds; review includes specific warnings. This is a screening aid rather than a guarantee, so multi-series figures should still combine color with markers, line styles, or direct labels. By default the catalog returns a lightweight summary and omits full HEX color arrays; pass include_colors=true when exact colors are needed.

Nature styling also applies non-color distinctions by default when a layer has multiple series: line charts cycle line styles, scatter charts cycle marker shapes, and line-symbol or polar charts use both. The response and diagnostics include series_distinction with the exact per-plot assignments and warn when the available patterns must repeat. Pass differentiate_series=false to origin_apply_nature_style when an existing template already owns these encodings. origin_apply_nature_style, origin_diagnose_graph, origin_plot_auto, origin_plot_chart_atlas, and FigureSpec style.palette_name can select a palette such as nature, lcpmgh_auto, or a local lcpmgh_006_001 style palette. The default remains nature, now backed by the local lcpmgh/colors Nature-style editorial palette. Use origin_palette_catalog(colors_count=6, family="lcpmgh/colors", include_colors=true) to list 6-color lcpmgh palettes, or origin_palette_catalog(min_colors=2, max_colors=16, family="lcpmgh/colors") to browse the local 2-16 color snapshot without returning every HEX value. Set ORIGIN_MCP_NATURE_PALETTE or ORIGIN_MCP_PALETTE to change the process-wide default.

Nature-style typography uses 20 pt as the minimum visual size across every output profile: legends, axis titles, tick labels, and general graph/image annotations are all at least 20 pt. FigureSpec annotations use the same 20 pt default unless style.annotation_font_size or an individual annotation style.font_size overrides it.

Nature styling resolves mark transparency before applying the preset. Dense scatter and line-symbol plots retain their data-driven transparency, while other Nature plots default to opaque marks. An explicit transparency passed to origin_apply_nature_style takes precedence, and graph diagnostics compare against the resolved value instead of assuming that every plot must be opaque.

origin_apply_nature_style(output_profile=...) also provides output-aware geometry presets. screen preserves the existing 20/18 pt interactive style; journal_single_column targets compact figures near 89 mm wide and journal_double_column targets figures near 183 mm wide, while retaining the project-wide 20 pt minimum and a 3 pt line width. presentation uses 24 pt axis titles, 20 pt supporting text, and 3.5 pt lines for projected slides. Explicit font sizes, line width, symbol size, and tick length override the selected profile. Nature-style line and scatter markers use a 10 pt default, and automatically generated numeric value labels use the same 20 pt minimum as other annotations. Nature-style diagnostics verify the resolved symbol size and transparency against the values Origin reports after applying the style, so a silently ignored Origin property assignment is surfaced as a warning.

Table plotting also applies readability defaults when the caller does not provide an explicit override. Machine-oriented headers such as measured_response, duration_s, and dose_uM are written to worksheet Long Name labels as Measured response, Duration (s), and Dose (µM). Common unit symbols use Origin Unicode escape notation so they survive export. A shared conservative rule resolver profiles chart family, series count, row count, category-label length, and numeric range. It hides redundant single-series legends, retains multi-series legends, moves line/scatter legends to the quieter upper corner, rotates crowded category ticks, selects scientific notation only for extreme magnitudes, keeps nonnegative bar-like charts on a zero baseline, and reduces marker size while increasing transparency for dense scatter plots. Continuous Cartesian axes request roughly five to seven major ticks according to label width and chart footprint, with a single minor subdivision. Datetime axes preserve Origin's date labels and use six temporal anchors; categorical axes preserve their category positions. Dual-Y plots format the two numeric scales independently, then use the smaller shared target tick count so the left and right major ticks stay visually aligned. A light horizontal major grid aids value comparison on line, scatter, bar, box, and generic Cartesian plots; vertical and minor grids remain hidden, and specialized heatmap, surface, and polar axes retain their own scale behavior. Standard Cartesian plots keep the top frame line but suppress duplicate top-axis major and minor ticks. Long numeric tick labels reserve additional left or right page margin automatically. Plots with long categorical X labels remain vertically oriented; their page height and bottom margin grow automatically so rotated labels are not clipped. The height is set from a target page aspect ratio, so repeated formatting does not keep increasing the page size. Axis titles are inferred from field semantics instead of concatenating every series name. Fields such as temperature_C, temperature_mean_C, and temperature_std_C produce the shared axis title Temperature (°C), while Mean and SD remain available as legend labels. Wide fields such as glucose_control_mg_dL and glucose_treated_mg_dL produce Glucose (mg/dL) with compact Control and Treated legend entries. Tidy value/unit data can also use a constant metric, measure, measurement, variable, or parameter column to recover the metric name and unit.

Dual-Y plots infer the left and right titles independently from the columns assigned to each axis. Explicit x_label, y_label, y1_label, and y2_label values always take precedence. For line and scatter charts, legend placement first estimates the legend footprint and checks all four inside corners against the plotted data. Each Y-axis group is normalized independently so dual-axis units do not distort the occupancy test. The quietest data-free corner is used when it fits. Only when every candidate intersects data, or the legend footprint is too large, does it move to a reserved column outside the plot; the page then widens and the right margin grows so the legend does not cover data or clip during export. Each graph response includes visual_defaults with the selected values and the reason for every choice. Explicit show_legend and palette_name arguments take precedence.

When style_mode="nature" and no palette is specified, table plotting uses lcpmgh_auto to choose an installed lcpmgh/colors palette whose color count matches the number of plotted series. Exact-count candidates that pass the accessibility preflight are preferred, then ranked by minimum and average OKLab separation, simulated color-vision separation, white-page contrast, and hue coverage. The selection notice exposes the same metrics. Dedicated table plotting tools and the parameterized origin_plot entry point accept palette_name; an explicit registered palette name always overrides automatic selection. Origin or custom template palettes remain untouched in origin_default mode.

Distribution and field-color tools use safer visual defaults: histograms choose a bounded Freedman-Diaconis bin width unless bin_width or a custom template is provided, and histogram/box/heatmap legends are hidden by default. Heatmaps use the perceptually uniform viridis colormap unless colormap or a custom template is supplied. Explicit arguments and FigureSpec plot styles always take precedence over these defaults.

For existing plots, origin_set_plot_style controls color, line width/style, symbols, transparency, column/bar width, colormaps, contour levels and color scale limits, histogram bin width, error-bar cap width, and box-chart width on a zero-based layer_index. Pass bar_gap to set Origin's -vg gap value; larger bar_gap values make columns or bars narrower. FigureSpec plot style entries can use the same fields for supported plot primitives.

FigureSpec plot group_style can apply safe per-series style sequences for multi-Y plots: colors, line_widths, bar_gaps, line_styles, symbol_kinds, symbol_sizes, transparencies, colormaps, contour/color scale sequences, and histogram/error-bar/box width sequences, or a series list of per-series style objects. FigureSpec uncertainty supports error-bar mappings that route to the existing safe plotting path, for example {"type": "errorbar", "y_error": "se"} or {"x_error": "xerr"}. It also supports worksheet-backed filled bands with lower/upper bound columns, for example {"type": "band", "lower": "lo", "upper": "hi", "transparency": 65}. Band support is intentionally narrow but covers multiple plot entries: each banded plot must have one named y column plus named x/lower/upper columns. The executor creates native Origin fill-area plots from contiguous x/lower/upper columns, overlays the main line, and trims generated legend text to the main series rather than the auxiliary band bounds.

For natural-language or registry-backed edits, origin_set_plot_property resolves a semantic property_name such as 柱宽, 折线粗细, 点大小, or 误差棒帽宽 against the plot style capability registry. It only applies properties that are marked implemented and map to a known safe route such as origin_set_plot_style, origin_apply_nature_style, or origin_apply_image_panel_style. Known-but-planned properties return applied=false with the matching capability and safe alternatives instead of guessing an Origin/LabTalk route.

Use origin_plot_style_setter_coverage(chart_type, plot_type_id) to audit whether registry entries marked implemented are executable through safe MCP routes. This is useful after adding a new style capability file or changing capability status.

Use origin_plot_style_capabilities(chart_type, plot_type_id, query) before changing an unfamiliar chart type. It is backed by the same registry as the knowledge base: core.json holds the small common capability set, while chart-specific JSON extensions such as column_bar.json, field_color.json, distribution.json, errorbar.json, image.json, three_d.json, area_pie.json, financial.json, and specialized.json are loaded only when the requested chart type, Plot Type ID, or query needs them. Every Plot Type ID in PLOT_TYPE_CATALOG has a style profile that maps it to one of these chart style families. The registry maps user-facing terms such as 柱宽, 折线粗细, 点大小, 色带, and 误差棒帽宽 to MCP setter parameters, Origin/LabTalk routes, readable fields, and implementation status. Properties marked implemented have stable MCP entry points; properties marked planned are intentionally documented so the assistant can report that a semantic setter is not yet available instead of guessing a LabTalk flag.

Extended Origin Workflows

The standard and full profiles expose the broader workflows added from the official originpro and X-Function surfaces:

• Data Connectors can be created from local or remote sources, inspected, updated, refreshed by sheet or project, expanded to another selection, and disconnected while keeping imported data. keep_dc=false removes the connector after the initial import.
• Matrix tools create/read/write matrix objects, set XY mapping and view controls, and transpose/rotate/flip them. Image tools import or create image pages, inspect/read/process them, and convert images to matrices.
• Analysis-template tools save/open OGW/OGWU templates. origin_batch_process covers file, folder, existing XY/XYZ, worksheet, and range inputs; origin_clone_import applies an existing connector/import structure to more files.
• Peak Analyzer tools expose theme/script/dialog execution, automatic baseline generation, and multi-Y batch analysis. The expert analysis surface includes typed FFT filtering, PCA, one-way ANOVA, and origin_run_xfunction, whose allow-listed catalog validates every argument before generating LabTalk.

Analysis results and recalculation

The analysis, standard, and full profiles expose a complete workflow for existing Origin analysis operations:

• Result reading through origin_get_analysis_results reads the report worksheet and uses Origin's official getresults X-Function to retrieve its result tree. The response provides stable parameters, metrics, sections, worksheet, and result_tree fields.
• Operation inspection through origin_get_analysis_operation reads a recalculating operation's settings tree through op_change op:=get.
• Recalculation through origin_recalculate_analysis runs the operation through op_change op:=run. Pass settings to replace the operation tree before recalculation, or omit it to use the current settings.

operation_range is the Origin range owned by the recalculating operation, for example [Book1]Result!col(2). Result-tree conversion uses the official originpro.utils tree helpers and cleans up temporary LabTalk trees after each call.

Advanced statistics

origin_multivariate_analysis provides typed dispatch for K-means clustering, hierarchical clustering, discriminant analysis, and partial least-squares regression. origin_nonparametric_test covers Friedman, two-sample Kolmogorov-Smirnov, Kruskal-Wallis, median, Mann-Whitney, paired sign, and one-sample or paired Wilcoxon tests. origin_survival_analysis covers Kaplan-Meier, Cox proportional-hazards, and censored Weibull workflows.

These tools accept worksheet columns or an explicit Origin range, validate all method-specific options against an allow-listed X-Function schema, and can send report output to output_book. They are available in the analysis, standard, and full profiles and require OriginPro where the underlying X-Function is OriginPro-only.

Publication multi-panel graphs

The plot, standard, and full profiles expose official graph-composition routes in addition to the existing FigureSpec and layer-arrangement tools:

• Graph merging through origin_merge_graphs combines graph pages into a new multi-panel graph with explicit rows, columns, gaps, margins, panel labels, and optional common scales. Common X-scale sizing works in Origin 2026; common Y-scale sizing uses resizeheightbyscale and requires Origin 2026b or newer.
• Layout creation through origin_create_graph_layout creates a Layout page containing selected graph pages while retaining them as separately editable graphs.
• Layer relationships through origin_link_graph_layers link X/Y scales and layer geometry; origin_copy_layer_scale copies axis-scale settings between layers.
• Layer extraction through origin_extract_graph_layers turns selected layers back into separate graph pages.

Use origin_merge_graphs for a single editable multi-layer figure and origin_create_graph_layout when the final page should assemble independent graph pages. Layer indexes in MCP calls are zero-based and are translated to Origin's one-based selectors.

Notes tools create, read, replace/append, load, export as HTML, and delete Notes windows. Syntax values are text, HTML, Markdown, or Origin rich text, with text or rendered view modes. Project-folder tools list, activate, create, move, and rename Project Explorer contents. Notes/folder deletion requires confirm=true; recursive folder deletion is explicit, the root folder is protected, and names/paths containing script delimiters are rejected.

Knowledge Base Tools

origin-mcp includes a local, structured knowledge base modeled as searchable and browsable collections. It is intentionally tool-addressable rather than only README text, so an MCP client can discover the right workflow before calling Origin.

General entry points are origin_browse_knowledge and origin_query_knowledge. In full profile, collection-specific browse/query helpers are also available, but the general tools are the stable default surface.

Collections:

• mcp_tools: origin-mcp tools grouped by workflow, such as worksheet, plotting, graph editing, analysis, export, and lifecycle control. Tool entries are generated from src/origin_mcp/tools/*.py docstrings so the index tracks the current MCP surface.
• reference: Origin workflow notes, Plot Type ID entries, style modes, graph formatting behavior, chart routing, analysis adapters, and runtime compatibility notes.
• python_api: OriginPro Python API usage notes used by this project.
• labtalk: LabTalk and X-Function routes used by this project, including worksheet plotting, plotxyz, plotm, legend refresh, export fallbacks, and analysis X-Functions.
• official_docs: versioned official OriginLab documentation boundary map for Python, originpro API class pages, LabTalk command/function/object references, and X-Function category references.

The local knowledge base is a curated operational index. It does not copy the entire OriginLab documentation set into this repository; entries that need exact official syntax include official_url, doc_family, doc_kind, versions, and verified metadata fields.

The official documentation boundary map has two layers: a hand-curated seed index in src/origin_mcp/knowledge.py and an optional generated overlay at src/origin_mcp/official_docs.generated.json. Refresh the overlay with:

python scripts\update_official_docs_index.py

The crawler follows OriginLab documentation links, classifies LabTalk command pages, X-Function pages, and originpro API class pages into stable browse paths, then validates duplicate paths and required version metadata before writing the JSON index.

Origin 2026b is the baseline index. Older supported versions use src/origin_mcp/official_docs.version_diffs.json, which stores only added, removed, and changed records for each version. At query time, origin-mcp applies that delta in memory so version="2024", version="2025", and version="2026" do not require duplicate full indexes.

To compare two generated indexes for Origin version drift:

python scripts\compare_official_docs_index.py old.json new.json --output diff.json

To build a compact version-diff overlay from separately generated version indexes:

python scripts\build_official_docs_version_diffs.py --base origin2026b.json --base-version 2026b --version-index 2026 origin2026.json --version-index 2025 origin2025.json --version-index 2024 origin2024.json --output src\origin_mcp\official_docs.version_diffs.json

Browse calls use a path-like topic. Examples:

{"collection": "reference", "topic": "plot-types/200"}

{"api": "originpro.find_graph"}

Query calls return ranked entries with path, title, summary, keywords, metadata, and score. Examples:

{"query": "heatmap plot type id", "collection": "reference", "limit": 5}

{"query": "legend font position", "limit": 3}

{"collection": "official_docs", "topic": "labtalk/commands/display-control", "version": "2026b"}

Single Source of Truth

Avoid maintaining hand-written exhaustive tool lists in docs. Use these knowledge queries instead:

{"query": "worksheet tools", "collection": "mcp_tools", "limit": 10}

{"query": "plotting recommended entry points", "collection": "reference", "limit": 5}

{"query": "legend font position", "collection": "reference", "limit": 5}

{"query": "analysis adapters include_output metrics", "collection": "reference", "limit": 5}

{"query": "runtime compatibility embedded bridge", "collection": "reference", "limit": 5}

The implementation remains the source of truth for schemas and callable functions. The knowledge base is the source of truth for searchable workflow guidance, Plot Type ID entries, style modes, graph formatting behavior, analysis adapter notes, runtime compatibility, and curated official documentation links.

See origin-bridge.md for bridge startup and real-Origin validation workflows. Use origin_doctor first for bridge issues.