talk

Communicate directly with iCog as a cognitive peer.

Transport: STDIO proxy Source: icog/mcp_runner.py

Signature

async def talk(message: str, current_task: str = "", scope_mode: str = "tiered") -> str:

Tool documentation

Communicate directly with iCog as a cognitive peer.

iCog recalls its memories about your current work, reasons over them, and responds — surfacing past decisions, asking important questions, or flagging patterns it's seen across sessions.

Use this to think out loud, get a second opinion, or let iCog interject with what it knows before you proceed.

Args: message: What you want to tell or ask iCog. current_task: One sentence describing what you're working on RIGHT NOW (e.g. "fixing the embed-500 bug in iCog backend" or "drafting the launch content kit"). PASS THIS EVERY CALL when the topic is non-obvious — without it, iCog falls back to vector similarity over past memories to guess your topic, which is the historical cause of it surfacing unrelated content. When you pivot topics, update this. iCog upserts it into your agent profile. scope_mode: How recall is scoped to this agent. "tiered" (default): in-scope memories first, out-of-scope appended with an explicit FROM-OTHER-PROJECTS label. "strict": only in-scope memories — use for brand-guard / production critique where cross-project contamination would be wrong. "lazy": legacy behavior (no scoping). Without an agent_slug (i.e. before identify), scoping is impossible regardless of mode.

Client usage

Call this tool through the hosted MCP endpoint at https://api.cognitivx.io/mcp/ with an authenticated MCP client. The local cogx CLI can also install the remote MCP server into supported agents.