AI & Agents MCP servers

Local-first memory brain for AI coding agents: hooks capture sessions, beliefs injected per prompt.

Local MCP proxy for tool restrictions, response redaction, audit logs, and database schema context.

Human oversight for AI agents: block a tool call on a verified human, with on-chain proof.

Deterministic, resumable multi-agent workflows across Codex, Claude, Cursor, and Kimi.

Attest agent decisions and verify records on a public ledger. Evidence, not just data.

Persistent memory for AI agents — audio intelligence + Music Memory ingestion and recall

Shared local knowledge vault MCP server for Claude Desktop (Chat + Code)

Orbit's local-first task and workflow registry for coding agents.

MCP server caching expensive LLM responses: HTTP-style freshness plus semantic recall.

Policy checks, approvals, records, and governed HTTP capabilities for unattended agents.

Run MCP tool calls through Runestone Gatekeeper — policy enforcement, human approval, audit.

Delegate a self-contained task to opencode's free-tier models as a one-shot subagent.

28-layer cognitive substrate with cross-session ToT evolutionary memory and self-telemetry tools

HYTHE — coordination bus + shared truth for AI agent fleets — self-hosted MCP server + stdio bridge.

Fully local memory for AI agents: zero-LLM ingest, semantic recall, air-gapped, auditable.

Local deterministic BM25 memory for AI agents. Offline-first, no API key, SHA-256 shards, stdio.

Value Context Protocol tools: VCP/I tokens, CSM1 codes, VCP-Lite, and context encoding.

Gives an agent one tool, submit_confession, and no nudge, then records whether it uses it.

Compatibility MCP server package that forwards to the correctover-mcp-gateway runtime.

Wrap any MCP server with enforced, bounded, revocable permissions — grant, narrow, check, revoke.

Transparent MCP proxy. Turns fat tool responses into queryable handles: 88-99% fewer tokens.

Verification and memory layer for coding-agent workflows.

Coordination rooms for AI coding agents: path locks, tasks, and conflict alerts. Metadata only.

Local checks for artifacts, declared claim evidence, and citation containment.

Outside-in validation gate for AI-generated text; a separate model checks LLM output, fail-closed.

Force evidence-gated, tiered claim reporting (FACT/INFERENCE/SPECULATION). No-LLM.

Spawn Claude Code instances from 6 platforms when 'Dear Claude' is mentioned.

Session handoff for Claude Code: list sessions and build handoff docs another agent continues from.