forkmind

Forkmind MCP Server

ForkMind records LLM calls as a local conversation graph for developers debugging prompts and agent workflows. Its MCP tools let agents inspect earlier attempts, trace branches, and archive or retrieve encrypted context capsules.

Server facts

forkmind MCP Server

Tech

Runtime / language
node
Transports
Not specified
Remote available
No
Authentication
Not specified

Activity

GitHub stars
3
Last commit
Data as of

Overview

Prompt experiments and agent debugging in ForkMind revolve around a locally stored graph of captured LLM calls. Developers can inspect requests, responses, tool calls, and token usage in a dashboard, then branch from an earlier turn with a changed prompt or model. Through MCP, agents can retrieve recent activity, inspect individual nodes, follow the path leading to a result, compare alternative continuations, and search recorded content. Tree statistics summarize the captured history. Capture uses a proxy, with SDK support and adapters for LangChain.js and the Vercel AI SDK.

A separate group of context tools lets agents move conversation material into encrypted capsules on disk. The workflow saves context, checks its digest and structure, and only then removes that material from the working window. Agents can browse capsules and restore either complete contents or selected segments after integrity checks. Other tools report storage statistics and replica status, synchronize replicas, and transfer capsules through passphrase-encrypted export and verified import. Explicit forgetting destroys the capsule key and marks its identifier as deleted.

Related MCP servers