ContextCord MCP Server
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@ContextCord MCP Serverresume my last session and show what changed since the checkpoint"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
ContextCord is a local-first continuity and decision layer for coding agents. It is not a chat-memory product: the repository remains the source of truth, old conclusions can be revalidated against current source, and the next session receives a bounded context packet instead of a transcript dump.
Quick start
git clone https://github.com/Yvniverse/contextcord.git
cd contextcord
python -m pip install -e .
contextcord init --profile generic
contextcord doctor
contextcord resume --assistresume --assist starts with a read-only continuation preview. A new source-bound session is created only after confirmation.
Related MCP server: 3Notch
Six modules, one local truth layer
Module | Problem it solves | What it owns | One-line value |
Continuum | How does engineering work survive a new session or host? | Task, Session, Checkpoint, Handoff, Closeout, Resume | Remember engineering progress, not chat history. |
Context Engine | What should the next agent actually see? | Bounded selection, BM25, freshness/authority signals, recheck, recoverable references | Select the context that matters instead of replaying everything. |
Veritas | Why should an old conclusion still be trusted? | Source identity, fingerprints, Evidence, Receipt chains, Qualification, replay | Keep why a conclusion was valid—not only the conclusion. |
Decision Plane | How are ambiguous choices and model routes decided? | Deterministic policy, Adaptive Router v3, Model Intelligence, receipts, Local Outcome | Models may advise; code keeps authority. |
MCP / Host Gateway | How can the same project continue across coding agents? | Dynamic MCP tool surface, Host Registry/configuration, BYOH adapters | One project state layer across multiple coding agents. |
Jev · optional | Which narrow decisions benefit from semantic judgment? | Optional typed provider on bounded decision paths | Use an LLM as a bounded semantic advisor, not an unrestricted executor. |
These are product modules, not six packages or databases. Internally, ContextCord uses one Capability Registry and one repository-local StateStore.
Use only what you need
contextcord feature profile minimal
contextcord feature profile continuity
contextcord feature profile decision
contextcord feature profile fullminimal— core + evidencecontinuity— continuity, memory, handoff, MCP/hosts and closeoutdecision— deterministic decision, Adaptive Router and Model Intelligencefull— all runtime capabilities, including Jev when installed/configuredcustom— explicit enable/disable combinations
Jev is intentionally separate from the default decision profile:
python -m pip install -e ".[jev]"
contextcord feature enable jev
contextcord provider jev statusProvider credentials stay local and are not written into public receipts.
How the pieces work together
Continuum ──→ Context Engine ──→ bounded continuation
│ │
└──────→ Veritas ───────────→ source-aware evidence
│
└────→ Decision Plane ──→ effective route
│
optional Jev
│
↓
MCP / Host Gateway
│
↓
real host execution
│
↓
verifier + local outcomeThe modules are a responsibility view, not a rigid one-file/one-module partition. Shared capabilities may contribute to more than one product module without creating a second registry.
Host integrations
Built-in integration contracts are provided for Codex, Cursor, Qoder, OpenCode and WorkBuddy, with a versioned BYOH adapter path for other hosts. “Built-in” describes the shipped integration contract; live model qualification is recorded separately in the current Host Evidence Registry.
Adaptive Router
The Decision Plane can choose an eligible model × reasoning-effort × role route. Hard code-owned rules define what can execute; external Model Intelligence is advisory; Jev is called only when a bounded semantic choice is useful. Only real host execution plus verifier output can become Local Outcome evidence.
See Adaptive Router and Research boundaries.
Architecture
The Architecture Atlas includes six complementary interactive Archify views: system architecture, Adaptive Router, workflow, sequence, data flow and lifecycle. The site shows six categories for the active language while retaining both Chinese and English standalone viewers.
Security and local data
ContextCord is local-first. Project state, provider credentials, raw Model Intelligence payloads and private delivery evidence are not part of the public repository surface. Host permissions and the operating system remain authoritative for real execution.
See SECURITY.md.
Project status
ContextCord is Public Alpha. Continuity, evidence, capability and MCP contracts are the primary product surface. Adaptive routing, external Model Intelligence and outcome calibration remain experimental behind explicit capability boundaries.
Contributing
Issues, documentation, reproducible benchmark methodology and carefully scoped host integrations are welcome. Start with CONTRIBUTING.md and follow the Code of Conduct.
License & acknowledgements
ContextCord 0.6.2a1 Public Alpha is distributed under the MIT License. Third-party and generated assets remain subject to their own notices and licenses; see THIRD_PARTY_NOTICES.md and ACKNOWLEDGEMENTS.md. Architecture documentation uses Archify; optional Model Intelligence may consume attributed public observations from CodexRadar / DRadar as an advisory data source.
This server cannot be deployed
Maintenance
Related MCP Connectors
Agent checkpoints. Resume after context resets and handoffs with retry-safe, versioned saves.
- OneLoreOAuthai.onelore
Shared project context for AI agents and teams: docs, tasks, and messages that stay current.
Project memory for coding agents: requirements, decisions, code graph and delivery telemetry.
Shared, versioned context that humans and AI agents can publish, review, annotate, and continue.
Related MCP Servers
- FlicenseNot gradedqualityCmaintenanceLocal-first deterministic project memory for AI coding agents, with context packs, decisions, gates, risks, scoped claims and explicit checkpoints in project-owned files.-
- AlicenseAqualityAmaintenanceLocal-first working-state checkpoints and portable handoff packets for AI coding agents. Continuation checkpoints recover Claude Code sessions after rate limits, crashes, and compaction; packets hand tasks across tools, repos, and machines — Markdown on disk, no cloud, no telemetry.2321 npm4MIT
- FlicenseNot gradedqualityBmaintenanceEnables coding agents to preserve and recover structured task state across context loss, recording objectives, constraints, decisions, and plan, and reconstructing a bounded deterministic context block on session resume.-
- AlicenseNot gradedqualityCmaintenanceEnables coding agents to persist and resume shared working state—tasks, progress, decisions, commands, tests, checkpoints, and handoffs—backed by SQLite while keeping Git authoritative for files and repository state.MIT