proposals

The operational memory + assembly line for competing in editais / RFPs / public tenders / incentive programs / awards. Multi-company (Revera / Itera / devfellowship + partners like B42): a company registry with a fiscal/legal profile, a governed answer library that composes across submissions by reference (never copy-paste), opportunities with an extracted requirements checklist, and submissions that assemble answers + vault documents for one applying company. Reads strategy/public, writes only its own proposals schema.


Endpoint	`https://proposals.mcp.devfellowship.com/mcp`
Tools	21
Backing data	`proposals` schema (companies + vault, answers, opportunities, submissions).

question_canonical is the question, and some funders ask for a budget, not for prose. proposals.answers and proposals.submission_sections therefore carry a sheet_id into work.sheets plus a kind discriminator (answer_kind / section_kind, one of markdown | sheet).

The order is always the same: create the spreadsheet on the engineering MCP (create_sheet, or copy_sheet to give one submission its own fillable copy of a template), then pass the new id here as sheet_id. The kind is inferred from sheet_id, so you rarely set it by hand; pass sheet_id: null to detach the sheet and go back to Markdown.

The sheet is referenced, never copied. harvest_answers promotes a sheet section into the library with the SAME sheet_id, so the library answer and the submission section stay one document.

ADR-5 of plan 20260909-dfl-sheets-primitive; columns from dfl-schema #956.

Companies + Document Vault

Tool	Description
`list_companies`	List the companies that can apply to an opportunity — own entities (Revera/Itera/devfellowship) and external partners (e.g. B42). Filter by relationship or search name/cnpj.
`get_company`	Get one company by id, optionally with its fiscal-year history (revenue/headcount) and contacts.
`upsert_company`	Create a new company, or update an existing one when `id` is given. `relationship` (own\|partner) is required when creating.
`upload_company_doc`	Upload a SMALL document into the vault by sending its bytes inline (base64): POSTs to the upload-file edge function as PRIVATE and records a company_documents metadata row. Body is capped (~10 MB body ≈ ~7.5 MB file) — for larger files use `link_company_doc`. Admin-gated.
`link_company_doc`	The definitive large-file path: record a vault row for a file ALREADY uploaded to the private bucket (upload-file edge function, visibility=private) — pass its media_id / media.devfellowship.com URL / share.devfellowship.com URL / storage path, no bytes in the body. Optional `supersedes` replaces a prior row (status=superseded + superseded_by). Admin-gated.
`list_expiring_documents`	The certificate-renewal radar: vault documents whose valid_until falls within a look-ahead window (already-expired included). Admin-gated.

Answer Library

Tool	Description
`search_answers`	Search the reusable answer library by free text and/or facet tags. The reuse entry point — find an approved answer, then reference it from a submission section. Each row reports `answer_kind` + `sheet_id`.
`create_answer`	Add a reusable answer to the library. company_id NULL = ecosystem-shared; set = company-specific. Optionally attach tags and pair a PT/EN translation. Pass `sheet_id` when the answer IS a spreadsheet.
`update_answer`	Edit an answer; optionally record an append-only revision snapshot (which carries `sheet_id` + `answer_kind` too) and/or fully replace its facet tags. Attach or detach a sheet with `sheet_id`.
`set_answer_status`	Transition an answer through its governance lifecycle: draft → approved → stale → retired. Approving stamps last_reviewed_at.
`list_stale_answers`	The library review queue: answers marked stale plus answers expiring within a look-ahead window. Each row reports `answer_kind` + `sheet_id`.

Opportunities

Tool	Description
`create_opportunity`	Register an edital/RFP/public tender/incentive program/award as a structured object (deadlines, value, status), optionally attaching a funder.
`add_requirements`	Bulk-add the extracted checklist for an opportunity (document/eligibility/content/form/budget), each with provenance back to the notice PDF.
`eligibility_check`	Cross a company against an opportunity’s requirements and return a per-requirement verdict + gap list (auto-checks documents against the vault and eligibility criteria against fiscal years).
`list_opportunities`	List opportunities in the pipeline, filterable by status/funder_id/kind — the discovery entry point before `create_opportunity` (avoid blind duplicates). Each row includes its resolved funder.
`search_opportunities`	Text search over title, notice_number (número do edital), or funder name — the dedup entry point before `create_opportunity`. Combine with status/kind/funder_id filters.

Submissions (Assembly)

Tool	Description
`create_submission`	Open a submission = one company applying to one opportunity (the multi-company anchor). Seeds the company as lead; pass `consortium` to add partners.
`upsert_section`	Create or update a section of a submission. answer_id is the reuse/provenance link; content_md is the opportunity-tailored final text adapted from that answer. Pass `sheet_id` when the section IS a spreadsheet.
`attach_document`	The compliance join: record that a specific vault document satisfies a submission — optionally tied to the exact requirement it fulfills.
`submission_checklist`	Assemble the full compliance + drafting state of a submission: every requirement with its attached documents, plus each section with its status, `section_kind` and `sheet_id`.
`harvest_answers`	The loop that compounds: promote reusable text written in a past submission back into the answer library, back-linking sections to the new answers. A sheet section promotes with its `sheet_id` intact.