SMI Aware Assistant

Fetches the full SMI record for a typed reference ("<type>:<id>") that search returned. Pass only an id that appeared in a search result; do not construct one. If you already have a plain SMI record id, use smi_get_deep_report or smi_get_export to read it, or smi_check_order_complete for its readiness.

Pre-connection onboarding helper, callable before any SMI token is set. With no input, returns the fork question "Do you already have an SMI token?" — call again with hasToken: "yes" for connect guidance (then verify_access), or hasToken: "no" for prospect-intake guidance. Runs locally; makes no network calls and never reads or requires credentials.

Open-ended discovery across SMI deep reports, research bundles, and exports when you do not already have a record id — finds candidates by subject name, url, or keyword and returns typed references to pass to fetch. If you already have a record id from the conversation rather than from a search result, call the specific tool instead: smi_get_deep_report or smi_get_export to read one record, smi_check_order_complete for order readiness, smi_list_deep_reports or smi_list_exports for filtered listings, smi_get_report_findings for findings within a report. Deep reports and research bundles are filtered server-side by subject first/last name, broadening to an unfiltered fetch ranked client-side (across id, name, url, summary) when the filtered fetch finds nothing; exports match on url.

Attaches a postal address to an SMI subject.

Attaches an alias to an SMI subject.

Attaches an educational institution to an SMI subject.

Attaches an email address to an SMI subject.

Attaches an employee to an SMI subject (for COMPANY-type subjects).

Attaches a url to an SMI subject employee, referenced by the subjectEmployeeId returned from smi_add_subject_employee.

Attaches an employer to an SMI subject.

Attaches a phone number to an SMI subject.

Attaches a relationship to an SMI subject, using a relationshipId from smi_list_relationships.

Attaches a url to an SMI subject relationship, referenced by the subjectRelationshipId returned from smi_add_subject_relationship.

Attaches a surname to an SMI subject.

Attaches a url to an SMI subject.

Scores free-text case notes against 5 weighted gap-analysis signals (no prior SMI order, high-relevance case type, known online presence, date of loss, social media mentioned) and returns a RECOMMEND/REVIEW/NO ACTION verdict, a confidence rating, which signals were/were not detected, a suggested product, and a best-effort orderData prefill (subject, matter id, case type, date of loss, handles/urls, suggested scope) for smi_create_deep_report. Runs entirely locally; makes no network calls.

Checks the status of a deep report or export record (by its record id, e.g. from smi_create_deep_report draftIds). Returns state = ready | in_progress | draft | failed | unknown and a terminal flag. Poll with bounded attempts and backoff while in_progress; stop as soon as terminal is true (ready or failed) — a failed record will never become ready. Also stop when state is draft: the record was never submitted, so it will not progress until smi_submit_deep_reports or smi_submit_exports is called on it.

Confirms a file already PUT to a presigned URL returned by smi_request_upload_url and persists the deep report upload record. Requires the version (the x-amz-version-id response header captured during the PUT).

Creates one or more SMI deep report drafts from subject orders. Returns the draft ids for review before submission. All orders in one call are filed under a single project; call once per project. A batch of up to 25 orders is created immediately. A larger batch runs as a background job that the server waits on for a few seconds, then returns the same draft ids plus the jobId. If the result has status queued, the job is still running or could not be read: call smi_get_job with its jobId to read the created deep report draft ids, and do not resend the orders. If the job failed, SMI does not say whether any draft was created: check existing drafts before resending anything.

Creates one or more SMI export drafts from profile-url orders. Returns the draft ids for review before submission. All orders in one call are filed under a single project; call once per project. A batch of up to 25 orders is created immediately. A larger batch runs as a background job that the server waits on for a few seconds, then returns the same draft ids plus the jobId. If the result has status queued, the job is still running or could not be read: call smi_get_job with its jobId to read the created export draft ids, and do not resend the orders. If the job failed, SMI does not say whether any draft was created: check existing drafts before resending anything.

Returns a consolidated, copy-paste-ready Markdown system prompt of SMI Aware house rules and a when-to-call-which-tool routing guide, plus exact paste-in locations for Claude Projects/Cowork/Desktop, Cursor, and other clients, so the assistant calls the right smi_* tools automatically. Runs locally; makes no network calls and takes no case data.

Fetches a single SMI deep report by id. Research Bundles are deep-report products, so use this tool for a Research Bundle id too rather than constructing a research_bundle: reference for the generic fetch tool. Accepts product-prefixed ids: RB-11751 (Research Bundle) and DR-11751 (Deep Report) both resolve here.

Lists releases for a given SMI deep report, with pagination. Each release lists its downloads: the report files (PDF, CSV, JSON) and the findings screenshot archive. Some releases carry only the findings archive and no report file; tell the user that in plain words. Use this tool to fetch the generated report PDF.

Fetches a single SMI export by id. Accepts product-prefixed ids: EX-30001 resolves here, while RB and DR ids belong to smi_get_deep_report.

Lists releases for a given SMI export, with pagination.

Reads the status of an SMI background job. Use it with the jobId that smi_submit_deep_reports, smi_submit_exports, smi_create_deep_report or smi_create_export return when a large batch is still running, to read its outcome. Returns status (pending | processing | completed | failed), a terminal flag, and failedIds with an error when the job failed. While a batch submission job is pending or processing, poll with bounded attempts and backoff, and stop as soon as terminal is true. A release job (type deep_report_release or export_release) stays pending until that report is released, which can take days: do not poll it in a loop. Tell the user they will be notified when the report is released.

Lists SMI release findings with pagination and optional filtering by report, platform, scope, and confirmation status. Report filters accept the displayed identifier (RB-11751, DR-402) or a bare numeric id, and platform filters accept a platform name (Facebook) or its numeric id. Each finding includes its summary, profile URL, and screenshot/evidence links. Findings are grouped by report and ordered by ascending sequence, so sequence 1 is Finding #1 of that report; refer to a finding by that number rather than by its id, which is an internal record key the reader cannot see. The ordering covers the returned page: SMI decides which findings a page carries, so a report with more findings than take may not begin at Finding #1 on page 1.

Reads a subject's full profile out of a deep report (no standalone GET /subjects/{id} exists upstream, so the input is a deepReportId, not a subject id). Returns the subject as structured JSON — any atomic field (e.g. date of birth via `subject.birthDate`) can be extracted from it directly.

Lists SMI deep reports with pagination and optional filtering.

Lists SMI exports with pagination and optional filtering.

Lists the SMI reference set of platforms.

Lists the SMI reference set of products.

Lists SMI projects with pagination.

Lists the SMI reference set of relationships.

Lists the SMI reference set of statuses.

Requests a presigned URL for uploading a large deep-report reference file (csv, doc, docx, gif, jpg, jpeg, pdf, png, txt, webp, xls, xlsx) directly to storage. This server never proxies file bytes: the caller must PUT the file to the returned signedUrl, capture the x-amz-version-id response header from that PUT, and then call smi_confirm_upload with that version to persist the upload record.

Submits previously created SMI deep report drafts (Draft status only) for processing. A batch of up to 25 ids is processed immediately and the result lists accepted and rejected ids. A larger batch runs as a background job that the server waits on for a few seconds, then returns the same accepted and rejected ids plus the jobId. If the result has status queued, the job is still running or could not be read: call smi_get_job with its jobId and do not resubmit the ids. If the job failed, SMI does not say whether any id was submitted: check each id with smi_check_order_complete before resubmitting. SMI notifies the user when each deep report is released; smi_check_order_complete reads the status of a single record.

Submits previously created SMI export drafts (Draft status only) for processing. A batch of up to 25 ids is processed immediately and the result lists accepted and rejected ids. A larger batch runs as a background job that the server waits on for a few seconds, then returns the same accepted and rejected ids plus the jobId. If the result has status queued, the job is still running or could not be read: call smi_get_job with its jobId and do not resubmit the ids. If the job failed, SMI does not say whether any id was submitted: check each id with smi_check_order_complete before resubmitting. SMI notifies the user when each export is released; smi_check_order_complete reads the status of a single record.

Partially updates an SMI subject via PATCH /subjects/:id. type (PERSON or COMPANY) is required on every call — the SMI backend requires it as a discriminator on every subject PATCH. Editable fields match the API contract: firstName/lastName (or companyName for a COMPANY), birthYear/birthMonth/birthDay, and gender. At least one editable field must be provided.

Submits a prospect contact request — first name, last name, email, and organization — to SMI so the sales team can follow up. For prospects who do not yet have an SMI Aware token.

Verifies the session Personal Access Token (PAT) against the SMI identity endpoint and returns the resolved account identity.