Licinexus MCP

What Licinexus MCP MCP server does

Licinexus MCP MCP server gives compatible AI clients structured access to Brazilian public procurement data from the Portal Nacional de Contratações Públicas (PNCP) and public company-registration data. It is suited to questions about open bids, contract history, winning suppliers, price-registration agreements, agency profiles, and what government bodies plan to buy.

The procurement tools cover searches and record-level lookups. An agent can find licitações, retrieve a complete procurement record, inspect its items, review results for a selected item, and list attached file metadata with direct URLs. Contract tools expose contract details, additive terms, and billing instruments. Supplier-oriented queries can list contracts associated with a CNPJ, while CNPJ lookup returns registration details such as legal and trade names, CNAE classifications, address, partners, capital, juridical nature, and Simples or MEI status.

How it works

The server is stdio-based. An MCP client starts the npm package and communicates with it through JSON-RPC over standard input and output; it is not intended to be run as an interactive terminal command. Requests are sent to public data endpoints, so no local database or API credential is needed.

PNCP searches require dates and, for licitações, at least one modality code. Defaults are provided when these values are omitted: recent searches generally use the last seven days and modalities 6, 8, and 9, while contract, price-registry, and PCA searches have their own documented default windows. PNCP accepts at most 365 days per request. Longer investigations must be split into separate windows. Keyword and value filters on some search operations are applied to the page returned by PNCP rather than across the entire result set.

Aggregation tools summarize procurement counts and optional values by period, and a comparison tool returns totals and absolute or percentage differences for two periods. Aggregation can make additional requests when filtering by government sphere or calculating value metrics, so large ranges and fine-grained buckets may require care.

Setup and configuration

Licinexus MCP MCP server requires Node.js 18 or later and an MCP-compatible client. The package is available as @licinexusbr/mcp and can be launched with npx -y @licinexusbr/mcp. Add that command and its arguments to the client’s MCP configuration, then fully restart the client so it loads the server.

The documented clients include Claude Desktop, Cursor, Continue, Cline, and programmatic MCP clients. ChatGPT consumer web does not natively support stdio MCP according to the supplied documentation, although an OpenAI Agents SDK setup is shown for programmatic use.

No API key, registration, or local database is required. CNPJ data uses BrasilAPI by default; setting CNPJ_PROVIDER=minhareceita switches the provider to Minha Receita.

Tools and capabilities

• Search licitações, contratos, atas de registro de preço, and PCA entries.
• Retrieve procurement, contract, price-registry, agency, and CNPJ details.
• Inspect procurement items, results, files, contract terms, and billing instruments.
• Find contracts associated with a supplier CNPJ.
• Aggregate procurement activity over time and compare two periods.
• Filter PCA entries by material or service classification and inspect planned items.

Limitations and notes

The server exposes metadata and direct URLs for licitação attachments; it does not download file contents. To inspect bidding results, the caller must first identify an item, normally by using the procurement-items tool. Price-registry lookups can include item balances and supplier information when the required procurement and registry identifiers are supplied.

PNCP enforces the 365-day request limit and returns HTTP 422 for wider individual windows. The aggregation tool supports a maximum overall range of about five years, but each underlying request still follows the PNCP limit. Client-side filters may only apply to the page fetched for a search, and aggregation with pagination can issue many upstream calls. These constraints should be reflected in agent prompts and workflows.

The project is open source under the MIT license. It is intended for MCP clients that can launch local stdio servers; a hosted remote endpoint is not described in the supplied material.