CognetivyCognetivy

MCP - any client

Standard protocol, zero lock-in

Cognetivy exposes a full MCP server via cognetivy mcp. Any MCP-compatible coding agent can connect to it - no proprietary format, no vendor lock-in.

Prerequisites

Install Cognetivy globally so the MCP server is available on your PATH:

npm install -g cognetivy

Then initialise a workspace in your project:

npx cognetivy

Connect your agent

Add the Cognetivy MCP server to your agent’s config. The exact location varies by client, but the server definition is always the same:

{
  "mcpServers": {
    "cognetivy": {
      "command": "cognetivy",
      "args": ["mcp"]
    }
  }
}

Workspace not in your project root?

Pass the workspace path explicitly:

"args": ["mcp", "--workspace", "/path/to/project"]

Works with any MCP-compatible agent

The installer also supports these agents directly via npx cognetivy:

OpenAI Codex

Add to Codex MCP config

GitHub Copilot

Add to Copilot MCP config

Gemini CLI

Add to Gemini CLI MCP config

Amp

Add to Amp MCP config

OpenCode

Add to OpenCode MCP config

CCR (Claude Code Router)

Add to CCR MCP config

And any other agent that speaks the MCP protocol - the server configuration above works universally.

Available MCP tools

Once connected, your agent has access to the full Cognetivy toolset:

workflow_get

Get the current workflow version

workflow_list

List all workflows in the workspace

workflow_set

Set a workflow version from JSON

run_start

Start a new run with an input file

run_status

Get run state and next step hint

run_step

Advance the run: start or complete a node

run_complete

Mark a run as completed

event_append

Append an event to a run

collection_get

Get collection items for a run + kind

collection_append

Append an item to a collection

collection_schema_get

Get the collection schema for a workflow

node_result_list

List node results for a run

Create your first workflowFull docs on GitHub