MCP Server

Cognova exposes a full Model Context Protocol server at https://cognova.dev/mcp. Any MCP-compatible client — Claude Code, Cursor, Windsurf, or custom clients — can connect to discover and invoke your workspace's agents, tasks, memories, and knowledge.

Access

All MCP tools are available on every plan, including Free. Agent invocation (ask_agent) consumes credits — rate limits are enforced per tier to prevent abuse.

Rate Limits

Rate limits are enforced per workspace, per minute:

Plan	Requests/min	Agent calls/min
Free	10	--
Starter	30	5
Pro	120	20
Team	300	60
Business	1,000	200

Setup

Claude Code

claude mcp add cognova https://cognova.dev/mcp \
  --transport http \
  --header "Authorization: Bearer cg_YOUR_KEY"

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "cognova": {
      "url": "https://cognova.dev/mcp",
      "headers": {
        "Authorization": "Bearer cg_YOUR_KEY"
      }
    }
  }
}

Other MCP Clients

Use the Streamable HTTP transport:

URL: https://cognova.dev/mcp
Auth: Authorization: Bearer cg_YOUR_KEY
Transport: Streamable HTTP (POST/GET/DELETE on single endpoint)

Available Tools

All tools are available on every plan.

Tool	Description
`cognova_list_agents`	List all workspace agents with name, description, model tier, and tools
`cognova_list_tasks`	List tasks with optional status and project filters
`cognova_get_task`	Get full task details by ID
`cognova_list_projects`	List all projects
`cognova_create_task`	Create a new task
`cognova_update_task`	Update task title, description, status, or priority
`cognova_complete_task`	Mark a task as done
`cognova_create_project`	Create a new project
`cognova_update_project`	Update an existing project
`cognova_recall`	Search memories by query with relevance ranking
`cognova_remember`	Store a memory in the workspace
`cognova_list_knowledge`	List all knowledge files
`cognova_get_knowledge`	Read a knowledge file by virtual path
`cognova_create_knowledge`	Create a new knowledge file
`cognova_update_knowledge`	Update an existing knowledge file's content
`cognova_list_secrets`	List all secret keys (values hidden)
`cognova_get_secret`	Get the decrypted value of a secret
`cognova_set_secret`	Create or update an encrypted secret
`cognova_get_balance`	Get current credit balance, plan tier, monthly credits, and reset date
`cognova_search_docs`	Search Cognova platform documentation
`cognova_ask_agent`	Send a message to a specific agent and get a response (uses credits)

Resources

MCP resources provide read-only data access (available on all plans):

URI	Description
`cognova://tasks/active`	All non-completed tasks as JSON
`cognova://knowledge/{path}`	Knowledge file content by virtual path
`cognova://agents/{agentId}`	Agent details (name, description, tools)
`cognova://docs`	Documentation table of contents
`cognova://docs/{slug}`	Individual documentation page (markdown)

Protocol Details

Cognova implements the Streamable HTTP transport (MCP spec 2025-06-18):

POST /mcp — Send JSON-RPC requests (initialize, tools/call, etc.)
GET /mcp — SSE stream for server-initiated messages (requires Mcp-Session-Id header)
DELETE /mcp — Tear down session

Sessions are maintained server-side with a 30-minute idle timeout.

Example: Raw JSON-RPC

# Initialize session
curl -X POST https://cognova.dev/mcp \
  -H "Authorization: Bearer cg_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "initialize",
    "params": {
      "protocolVersion": "2025-06-18",
      "capabilities": {},
      "clientInfo": { "name": "example", "version": "1.0.0" }
    }
  }'

# List tools (use Mcp-Session-Id from init response)
curl -X POST https://cognova.dev/mcp \
  -H "Authorization: Bearer cg_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -H "Mcp-Session-Id: SESSION_ID" \
  -d '{
    "jsonrpc": "2.0",
    "id": 2,
    "method": "tools/list"
  }'