loopio

Working with Loopio

Requirements

This skill drives the tools of the loopio-mcp MCP server (search_library, get_library_entry, get_library_structure, list_projects, get_project, get_project_questions, get_project_status_summary, and the gated write tools). If these tools are absent, the server is not configured or the tool's tier is disabled; say so and point to https://github.com/fredericboyer/loopio-mcp#readme. Never simulate a tool result or fabricate library content.

Domain model

• Library: stacks contain categories, which contain subcategories. Entries live at a location: {stackID, categoryID?, subCategoryID?}.
• Entry: one or more question phrasings sharing a single answer, plus tags, language, and review status. Entries are the reusable answer record.
• Project: an RFP/RFI/DDQ/questionnaire being answered. Projects have sections and subsections containing project entries (question, current answer, status). Project entries are per-project; library entries are the shared source of truth.
• Resolve names to IDs with get_library_structure (no arguments); never guess a stackID or categoryID.

Searching well

• Query the question's distinctive nouns, not the whole sentence: for "Describe how your organization encrypts data at rest", search encryption at rest, not the full question.
• Scope with locations (IDs from get_library_structure) when the library is large or the term is ambiguous.
• Prefer fresh content: filter with lastUpdatedDate: { gte: "<ISO-8601>" } or sort candidates by their last-updated date.
• Targeted matching: searchInQuestions / searchInAnswers / searchInTags narrow where the query applies; exactPhrase for verbatim strings; synonyms to broaden.
• Results are capped. truncated: true means there were more matches than returned: narrow the query (more specific terms, a location, a date filter) instead of asking for everything.
• No hits? Broaden stepwise: drop a filter, then try synonyms or alternate terms (e.g. "SSO" / "single sign-on" / "SAML"), then search tags.

Answering RFP and questionnaire questions

1. Search the library for each question (terms as above).
2. Evaluate candidates: relevance first, then freshness and review status. Pull the full text with get_library_entry (id) before relying on an entry; expandMergeVariables: true substitutes placeholder variables.
3. Draft the answer grounded in the best entries: adapt tone and wording to the question, but every factual claim must come from a library entry.
4. Cite which entry IDs the answer drew from, so a human can verify.
5. No supporting entries? Report it as a library gap. If a draft is still useful, clearly mark it as ungrounded and needing human review. Never invent capabilities, certifications, numbers, or policy claims.
6. Inside a project: get_project_questions (projectId, optional sectionId/subSectionId) lists what needs answering; answer_project_entry (id, answerText) writes an answer back (write-gated, see below).

Library curation

• Search before you create. Always run search_library for the new question's terms first; if an equivalent entry exists, update it instead of duplicating.
• Create with create_library_entry: questions (one or more phrasings), answerText, location (stackID required), optional languageCode and tags.
• Update with update_library_entry (id + JSON Patch), e.g. replace an answer: { "op": "replace", "path": "/answer/text", "value": "..." }.
• Maintenance loop (run periodically): report → identify stale → review → update:
  1. get_project_status_summary (lastUpdatedDateGt) shows recent project activity; search_library with lastUpdatedDate: { lte: "<cutoff>" } finds entries not touched since the cutoff.
  2. Present stale candidates to the user with age and content summary.
  3. Update approved revisions via update_library_entry.
• delete_library_entry is permanent. Confirm with the user before any delete, and prefer updating over deleting.

Write and delete gating

Mutating tools are opt-in on the server: create_library_entry, update_library_entry, and answer_project_entry require LOOPIO_ENABLE_WRITES=true; delete_library_entry additionally requires LOOPIO_ENABLE_DELETES=true. Gated-off tools are not registered at all, so their absence is configuration, not a bug. If asked to write while the tools are missing, explain the gate instead of retrying.

Projects and reporting

• list_projects filters by rfxTypes (RFP, RFI, DDQ, SQ, PP, OTHER) and owners (user IDs).
• get_project (id) for one project's data; get_project_questions for its entries.
• get_project_status_summary (lastUpdatedDateGt, ISO-8601) summarizes project status for reporting and triage; it is also the entry point of the maintenance loop above.

Troubleshooting

• 401 / auth errors: credential problem (LOOPIO_CLIENT_ID / LOOPIO_CLIENT_SECRET, or the Loopio app lacks a needed scope).
• Tools missing entirely: server not configured in the MCP client.
• Only write/delete tools missing: gating (see above), by design.
• truncated: true: result cap reached; narrow the query.
• Server config reference (env vars, scopes, setup): https://github.com/fredericboyer/loopio-mcp#readme