AgentContextKit MCP server
Commit Oct 15, 2026Prepares repositories for AI agents by scoring readiness, optimizing context packs, and enforcing evidence-backed task completion workflows.
Also known as Cynrath/agent-context-kit
- Stars
- 14
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Oct 15, 2026
- License
- MIT
What is AgentContextKit MCP?
AgentContextKit prepares repositories for AI agents by scoring readiness, optimizing context packs, and enforcing evidence-backed task completion workflows.
Summary generated from the project's README
About AgentContextKit MCP server
Preparing codebases for AI agents is what agent-context-kit enables for developers using tools like Claude Code, Cursor, or GitHub Copilot. It acts as a deterministic engineering layer that scores repository readiness, optimizes context efficiency, and enforces strict task completion workflows. By running locally on Node.js 22 or higher, it provides an offline-first experience that ensures no code or secrets leave your machine during analysis.
Users can utilize the ackit CLI to scan for security risks like connection strings or absolute paths and to identify conflicting instructions across files like AGENTS.md or CLAUDE.md. The tool generates provider-aware context packs tailored for specific models such as Gemini, Codex, or Copilot, helping to reduce token waste. It also implements a trust flow that requires independent verification and evidence-backed verdicts before a task can be marked as complete, preventing unverified changes from being merged.
What you can do with it
- 1Optimize AI context usageReduce the number of tokens wasted on duplicate guidance or oversized files.
- 2Enforce verified task completionEnsure that AI-driven tasks are only completed after providing independent proof of success.
- 3Secure sensitive repository dataScan your files to prevent secrets and private keys from being included in AI prompts.
Written from the project's README. Every use case is checked against a quote from it.
Automated checks
Ran Nov 14, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachablecynrath/agent-context-kit is public and reachablegithub.com/cynrath/agent-context-kit ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-10-15github.com/cynrath/agent-context-kit ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubACKit — AgentContextKit
npx --yes @cynrath/agent-context-kit@latest readiness
- Detect conflicting agent instructions (
AGENTS.md/CLAUDE.md/ provider scopes) - Find unsafe repository context (secrets, connection strings, absolute paths)
- Measure repository readiness across instructions, security and context efficiency
- Reduce wasted agent context (duplicate guidance, oversized files, low-value includes)
- Enforce evidence-backed task completion with independent verification
- Generate provider-aware context packs (Codex / Claude / Copilot / Gemini / generic)
✨ Demo (30 seconds)
# 1 — onboard (dry-run, writes nothing)
$ ackit init --dry-run
✔ plan: AGENTS.md shim + 4 built-in skills
# 2 — score readiness (explainable, no LLM)
$ ackit readiness
Readiness 88/100 ██████████████████░░ (threshold 80 — pass)
Instructions 90/100
Security 90/100
Context Efficiency 70/100
# 3 — graph + optimize
$ ackit instructions --explain
codex/AGENTS.md → claude/CLAUDE.md (provenance: frontmatter applyTo)
$ ackit optimize --explain --json | jq .suggestions[0].tokenWasteEstimate
42
# 4 — provider-aware pack & dashboard
$ ackit pack --profile codex --max-tokens 50000 | head -5
# ACKit Context Pack — 12 files, 34210 tokens
$ ackit dashboard --port 0 --open # localhost-only, CSP, live polling
→ http://127.0.0.1:54321
🔐 Trust flow (60 seconds)
$ ackit task complete TASK-0007
completion gate blocked: MISSING_VERIFIER_VERDICT: profile 'standard'
requires an independent verdict (run 'ackit verification bundle' + record)
$ ackit evidence verify TASK-0007 --criterion AC-001 --type test --ref "pnpm vitest run (green)"
$ ackit verification bundle TASK-0007 --format json --out .ackit/reviews/bundle.json
$ ackit verification record TASK-0007 --verdict .ackit/reviews/verdict.yaml --bundle .ackit/reviews/bundle.json
TASK-0007: verdict VR-0001 registered (PASS)
$ ackit status TASK-0007
blockers: (none — completion eligible)
- verdict VR-0001 (PASS): fresh; independent
$ ackit task complete TASK-0007
TASK-0007: completed
$ ackit checkpoint export TASK-0007 --format json --out .ackit/reviews/handoff.json
# ... second process, zero shared memory ...
$ ackit checkpoint import .ackit/reviews/handoff.json
TASK-0007: handoff CP-0001 fresh (bundle 9f2c…); resume context renders
Stale state blocks with VERDICT-STATE-STALE (naming what changed);
a fresh bundle + verdict restores eligibility. The full flow runs as a
test (tests/e2e/trust-flow-demo.test.ts) and is walked through in
docs/guides/demo-trust-flow.md —
offline, no API key.
🧭 Table of Contents
- Why ACKit?
- Features at a glance
- Install
- Quickstart
- CLI Overview
- Configuration
- Architecture
- GitHub Action
- VS Code
- Docs
- Development
- Versioning
❓ Why
| Before | With ACKit |
|---|---|
| Convention-based — nothing is verified | Deterministic local analysis with stable JSON/SARIF contracts |
| Cloud-coupled — code leaves the machine | Offline-by-construction — zero network in product code |
| Secrets leak into prompts/logs | Redacted at construction — evidence never contains plaintext secrets |
One AGENTS.md for all providers |
Provider-aware — Codex/Claude/Copilot/Gemini/generic, includeScopes/applyTo |
| “It worked on my machine” | Machine-checkable gates — ackit scan --ci, ackit readiness --strict |
This repository dogfoods itself: it was built by following the same
docs/taskstask system it ships.
What ACKit is not: an autonomous coding agent, a model runtime, a browser automation framework, an agent router, or a cloud orchestration service. ACKit is the deterministic, offline-first engineering layer those agents share — repository context, explicit intent, structured tasks, plan discipline, evidence-backed completion, resumability, independent verification, policy enforcement, drift detection, and provider interoperability. Model execution and autonomous coding stay with your chosen agent (Codex, Claude Code, OpenCode, Copilot, Gemini, Cursor, Cline, any MCP-capable agent).
🎁 Features at a glance
📦 Install
Requires Node ≥ 22.
# global (recommended)
npm install --global @cynrath/agent-context-kit
ackit --version # 0.5.4
# one-shot, pinned
npx --yes @cynrath/agent-context-kit@0.5.4 --version
npx --yes @cynrath/agent-context-kit@0.5.4 --help
# from source
pnpm install --frozen-lockfile && pnpm build
node dist/cli/index.js --help
🚀 Quickstart
ackit init --dry-run # plan shims + 4 built-in skills (writes nothing)
ackit scan --ci # gate: exit 1 at/over threshold (medium)
ackit readiness # 0–100, N/A renormalization, --strict/--fail-below
ackit instructions --explain # graph v2 with provenance
ackit optimize --explain # 8-class advisor, waste estimates
ackit pack --profile codex --max-tokens 50000 # provider-aware, budgeted
ackit diagnostics --json | jq .profile
ackit dashboard --port 0 --open # localhost-only
Every command supports
--jsonand--help. Full tour →docs/guides/getting-started.md
AgentContextKit MCP questions
Does it require an internet connection or API keys?
No, it is offline-by-construction and requires no network or API keys for its core functionality.
Can it help reduce token usage for AI agents?
Yes, it can measure and optimize repository context to reduce wasted agent context like duplicate guidance or oversized files.
Is it compatible with different AI providers?
Yes, it generates provider-aware context packs for Codex, Claude, Copilot, Gemini, and more.