CLEATUS MCP Server

Connect Claude, Claude Code, ChatGPT, and any other Model Context Protocol (MCP) client to the CLEATUS platform. Ask an AI assistant to search federal opportunities, manage your pipeline, create tasks, and run workflows — directly against your live CLEATUS account.

Server description

CLEATUS is the platform for government contractors: a unified workspace for discovering contract opportunities, qualifying pursuits, and winning work with federal, state, and local agencies. Our MCP server exposes that workspace as structured tools an AI assistant can call on behalf of the signed-in user — scoped to a single entity per OAuth token, with the same permission model as the web app.

Features

Discover

Search 1M+ federal contract opportunities (SAM.gov, GSA, SeaPort, agency forecasts). Get AI-ranked recommendations scoped to your entity's NAICS, capabilities, and past performance.

Evaluate

Semantic market intelligence across awards, agencies, contractors, people, and NAICS classifications — powered by BGE-M3 embeddings and a domain reranker.

Pursue & manage

Add opportunities to your pipeline, create and assign tasks, manage saved searches, organize documents, and trigger automation workflows — every mutation audit-logged the same way the web UI logs them.

Context

Each tool call is scoped to the entity attached to the OAuth token. Notifications, team members, and credit balance are all surfaced to the assistant so it can answer questions without leaving the conversation.

Setup

Claude.ai

Open the Claude web app → Settings → Connectors → Add custom connector. Paste https://www.cleat.ai/api/mcp/mcp as the server URL and follow the OAuth login.

Claude Code

Add to your project's .mcp.json:

{
	"mcpServers": {
		"cleatus": {
			"command": "npx",
			"args": ["-y", "mcp-remote@latest", "https://www.cleat.ai/api/mcp/mcp"]
		}
	}
}

ChatGPT / custom GPTs

Point any MCP-capable client at the endpoint. PKCE S256 is required; see the authentication section below.

Authentication

The MCP server implements OAuth 2.1 authorization code flow with PKCE (S256). Discovery metadata is available at /.well-known/oauth-authorization-server, following RFC 8414.

• Authorize: /oauth/authorize — Clerk-backed login with an entity picker. The user selects which CLEATUS entity the token is scoped to.
• Token exchange: /oauth/token — exchanges an authorization code plus PKCE verifier for an access token and refresh token. Refresh rotation is supported.
• Scopes: read and write. v1 tokens receive full access to the chosen entity; granular scope enforcement is on the v2 roadmap.
• Rate limits: 100 reads/min and 30 writes/min per access token. Exceeding returns HTTP 429 with a Retry-After indication.

Partners building a listed connector should contact us to have a client registered with allowlisted redirect URIs.

Usage examples

Example 1 — Find active opportunities (read)

User prompt: "Find IT services opportunities under NAICS 541512 posted this week."

Tool called: cleatus_search_opportunities

Arguments:

{
	"naics_codes": ["541512"],
	"posted_since": "2026-04-10T00:00:00Z",
	"status": "active",
	"limit": 20
}

Response: markdown table of the matching contracts with title, agency, NAICS, set-aside, posted/deadline dates, match score, and a view_url deep-link for each row. Pagination via cursor when has_more is true.

Example 2 — Add to pipeline and assign a task (write)

User prompt: "Add that second result to my pipeline and assign a proposal kickoff task to Sarah due next Friday."

Tools called: cleatus_add_to_pipeline then cleatus_list_team_members (to resolve "Sarah" to a user id) then cleatus_create_task.

Result: a new pursuit on the entity's kanban, plus a task titled "Proposal kickoff" with due date and assignee. Sarah receives the same in-app notification and email she would from the web UI. Both actions are audit-logged with the originating OAuth client for traceability.

Example 3 — Kick off an automation workflow (complex)

User prompt: "Run the FAR compliance analysis workflow on the Navy solicitation I just added."

Tools called: cleatus_list_workflows (locate the workflow) then cleatus_get_pursuit (confirm the target) then cleatus_run_workflow.

Result: { run_id, status: "queued", view_url }. The assistant can share the view_url so the user can monitor progress; the workflow executes the same nodes (AI extract, agent review, document generation) it would from the web UI.

Data handling & privacy

• Every call is scoped to one entity via the OAuth token. One token cannot read or modify another entity's data.
• We do not log prompts or tool-call arguments. Server-side logs capture tool name, token id (hashed), success/failure, and timing.
• Credit-consuming tools (currently only cleatus_search_market at 1 credit per call) debit the same wallet as the web app.
• Full privacy policy: https://www.cleat.ai/privacy.

Support

• Email: support@cleat.ai
• Security disclosures: security@cleat.ai
• Security & compliance: https://www.cleat.ai/security