Contextd

Contextd is undergoing a complete re-write. The goal is to keep the focus squarely on context sharing between human engineering teams, their agents, and autonomous agent workflows — regardless of platform.

The old contextd was single-agent memory that grew into ~24 MCP tools and a pile of context-window mechanics: checkpoints, folding, repo search, conversation indexing, reflection, temporal workflows. None of that was team knowledge. All of it is cut. This is a clean break — no migration, no compatibility shim, no data import. The idea survives; the code doesn't.

What replaces it is a server. Knowledge lives in one place, is written and read by humans and agents on equal footing, and is delivered back into whatever tool the engineer happens to be using.

Build institutional knowledge across the Org, its Teams, and its Projects.

Everything is a Record, and every Record belongs to one of three families. Incidents capture what broke and what fixed it — a Problem (the error signature, the symptom, the environment) and any number of Remediations attached to it, because the same error legitimately has different correct fixes in different contexts. A Problem with no fix yet is still knowledge worth keeping. Decisions capture a choice and its rationale, covering both "we require two reviewers" and ADR-grade architecture. Standards are standing constraints that apply everywhere.

Records attach to a point on the ladder — Org → Team → Project — and a Project's identity is stable across every branch and worktree, so knowledge earned on a feature branch is not second-class. Individual contributors aren't a level on that ladder; they're the authors. Every Record carries the Principal who wrote it, so "who knew this" is answerable without carving out a private tier that would defeat the point.

Keep it accessible via RAG, so departures stop costing you the knowledge.

An engineer leaves? No problem. If they used an LLM and contextd, their troubleshooting, their technical decisions, and the reasoning behind them stay behind for the rest of the team to learn from. Search cascades upward — a query from a Project sees the Project, its Team, and the Org, merged and re-ranked with a bias toward the narrower scope rather than a narrowest-first fallback that misses the excellent Org-level answer sitting one level up. Records are immutable; a correction creates a new linked Version, so it's always answerable what the guidance said on the day someone followed it. Confidence is earned by corroboration — the same Problem recurring in another Project, the same Remediation working twice — never by who or what wrote it.

Let autonomous workflows draw on the same context your engineers build.

Agents are first-class Principals with their own identity and their own permissions — not humans-by-proxy, not bounded by whoever launched them. That means an agent's writes are attributable to it, revocable independently, and auditable on their own terms. It also means the shared knowledge base has one substrate, not a human half and a machine half that drift apart. Nothing enters that shared base without crossing the promotion gate: Incidents can auto-promote once confidence clears a threshold, Decisions and Standards require a human to approve. The gate is the trust boundary, and it's also where secrets get caught on the way in.

Over time the Incidents feed synthesis — clusters of related Incidents propose a Standard, a human approves it, and the knowledge base improves itself.

Define guardrail policies that reach every agent in the Org.

Standards are what makes this a system of record rather than a wiki nobody reads. They are never searched for — they are injected, unconditionally, into every agent session at startup, whether the agent thought to ask or not. That's the difference between a constraint and a suggestion. A thin local client is the only thing that runs on a developer machine: it fetches active Standards and injects them. Not a daemon, not a sync protocol, no local knowledge base.

Platform independence is a property of that design, not an aspiration.

Claude Code, Cursor and Codex converged on the same session-start hook shape, so one small binary covers all three. Hosts without hooks — VS Code, Gemini CLI — get the same Standards compiled to disk as a CLAUDE.md / AGENTS.md fragment, which has the side benefit of being legible to an engineer who isn't running an agent at all.

When a Project genuinely can't satisfy an Org Standard, the answer is an Override: a narrower-scoped Record that wins for that scope only, without retiring the Standard everywhere else. Exceptions are recorded and approved, not quietly ignored.

Chia Health MCP Server — Patient workflow integration for a licensed US telehealth platform. Browse GLP-1 medications (semaglutide, tirzepatide), peptide therapies (sermorelin, NAD+, glutathione), and longevity treatments. Check eligibility, complete intake, sign consents, and manage treatment plans. 30 tools, HIPAA-compliant. All prescriptions evaluated by licensed US healthcare providers and delivered from FDA-regulated pharmacies across 50 states + DC.

Broker + MCP server for last-bidder-wins games on Solana — agents register, auto-fund a Privy wallet, and bid via streamable HTTP

AI-powered no-code app builder with 17 MCP tools — create projects, generate pages from natural language, AI text/image generation (GPT, Claude, Gemini, 14+ models), page CRUD, workflow execution, publish & version control. SSE transport, API key auth.

An mcp server for your food ordering needs.

Agent-to-Agent handoff certification for multi-agent systems — validates context preservation, verifies agent capabilities before handoff, logs transfer chains, and ensures no data loss in agent orchestration.

Unified MCP & skill management gateway with progressive disclosure. Manages multiple MCP servers as Agent Apps, loading tool schemas on demand for 99% context token savings. Shared across Claude Code, Codex, OpenCode and more.

A collection of Model Context Protocol (MCP) servers for various tasks and integrations, supporting both Python and Node.js environments.

Open-souSecurely feeds real security refreshed rules into Cursor, Claude Code, and Windsurf — zero config, no API key.

Health intelligence MCP — access biomarkers, biological age, and personalized longevity action plans from your Aniva profile.

Real-time stock heatmaps and investment tools delivered as interactive React components.