Axon
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., "@Axonlocalize the bug causing the crash in user auth"
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.
Axon
Root-cause debugging & vulnerability MCP server and CLI for AI coding agents.
Axon traces a fault signal back to its source — like the axon that carries a neuron's signal. It gives an agent the tools to localize bugs, verify fixes, and find security vulnerabilities with high precision.
Companion to Cortex. Cortex is the context/graph brain; Axon is the debugging + security tracer. Cortex is optional — Axon runs fully standalone.
Current release facts (2026-07-09 docs refresh): package/plugin version
0.5.11; published distribution name axon-debug; executable name axon.
Why Axon
Research shows no agent hits "90% detect everything" universally. Axon reaches ≥90% on a defined slice by stacking four proven levers:
Context grounding — code graph + retrieval, built-in (no external deps).
Executable verification — reproduce → fix → confirm FAIL→PASS.
Adversarial triage — a refuter agent must disprove each finding → kills false positives (the security adoption killer).
Static-analysis grounding — SAST output the agent filters/confirms.
Related MCP server: yo-bug
Plug-and-play by design
Zero-config. Install → run. Nothing else needed.
No GPU, no model download, no daemon. Retrieval defaults to BM25 + ripgrep.
Dependency-light startup. Plugin startup uses
bin/axon-mcp.py, which importsaxon.mcp_stdiodirectly; no plugin-localpip installor venv bootstrap is required.Optional accelerators, auto-detected, never required:
Cortex → richer code graph/impact. Preferred transport is Cortex's MCP server (
cortex mcp; override withAXON_CORTEX_MCP_CMD, set it tooffto force the CLI), which reuses Cortex's persistent per-repo index. Falls back to thecortexCLI (per-call budgets tunable viaAXON_CORTEX_INGEST_TIMEOUT/_BUNDLE_TIMEOUT/_GRAPH_TIMEOUT), then to the built-in provider — logging why it degraded.mcpextra → FastMCP-based server path when you want it.semgrepextra → Semgrep-backed SAST when you want it.
Same agent tools and output shape on every rung. Accelerators only improve quality/speed; their absence never breaks a feature.
Tools (MCP)
axon serve exposes Axon's MCP registry over stdio, including index,
graph_context, search, status, localize, run_tests, repro,
verify_fix, rank_patches, spectrum, inspect, investigate,
sast_scan, refute, and triage.
Quickstart
pip install -e . # or: uvx axon-debug
axon --help # current subcommands: serve, index, doctor, gc
axon doctor # environment + active backend
axon index <repo> # build/refresh the code index
axon serve # start the MCP server (stdio)Point any MCP host (Claude Code, etc.) at axon serve. Runs with no Cortex,
no GPU, no model download.
Install as a Claude Code plugin
Inside an interactive Claude Code session, use slash commands:
/plugin marketplace add "/path/to/Axon"
/plugin install axon@axonFrom a shell, use the Claude Code CLI equivalents:
claude plugin marketplace add "/path/to/Axon"
claude plugin install axon@axonYou can substitute the Git remote if you prefer not to use the local clone path:
https://github.com/alilfrances/Axon.git
Install as a Codex plugin
codex plugin marketplace add "/path/to/Axon"
codex plugin add axon@axonCodex can use the same Git remote source instead of the local path:
https://github.com/alilfrances/Axon.git
Official command references: Codex plugin marketplace CLI, Codex plugin install flow, and Claude Code plugin marketplaces.
Start a new Claude Code or Codex session after installation so the plugin MCP server and hooks are loaded.
Plugin startup needs python3 and does not run pip or create a venv.
bin/axon-mcp.py loads the dependency-free axon.mcp_stdio adapter directly.
Semgrep-backed SAST is optional: install semgrep on PATH to enable it.
Codex installs plugins into its local plugin cache; after updating Axon, run
codex plugin marketplace upgrade axon and restart Codex so the MCP server is
loaded from the refreshed cache.
Versioning: Axon uses semver; bump pyproject.toml,
src/axon/__init__.py, .claude-plugin/plugin.json,
.claude-plugin/marketplace.json, and .codex-plugin/plugin.json together on
release.
Status
Current package/plugin version is 0.5.11. The repository's pytest suite now
contains 115 test functions. Older v0.3 / 60 tests statements are retained
only in dated planning or verification records.
See docs/VERIFICATION.md for the dated verification
log and docs/PLAN.md for the original planning document.
v0.3 debugging upgrades — function-level localization, runtime-state inspection, patch ranking, and one-call investigate bundles add deterministic evidence for root-cause debugging workflows. Measured on the django/sympy SWE-bench Verified slice (n=8, no LLM, no regression from v0.2): File@3 50%, File@10 75%, Function@10 12% (new deterministic-only baseline — the calling agent is the reranking layer).
Security
Report vulnerabilities privately through the process in SECURITY.md.
Public repository hardening requirements are tracked in
docs/GITHUB_SECURITY.md.
This server cannot be deployed
Maintenance
Related MCP Connectors
- JamOAuthdev.jam.mcp
The Jam MCP server provides AI tools with instant bug context without manual prompting, enabling a streamlined workflow from bug identification to ticket creation and pull request generation without switching between tools.
MCP server for building and testing AI agents with multi-model experimentation and insights.
Zero-config MCP security scanner for AI-generated apps. 25K+ vulnerability patterns.
ArcAgent MCP server for bounty discovery, workspace execution, and verified coding submissions.
Related MCP Servers
- AlicenseBqualityDmaintenanceAn MCP server that enhances AI agents' coding capabilities by providing zero hallucinations, improved code quality, security-first approach, high test coverage, and efficient context management.159 npm1MIT
- AlicenseNot gradedqualityDmaintenanceAn MCP server that gives AI coding assistants QA superpowers, enabling users to report bugs by pointing, clicking, or typing while automatically capturing diagnostic data for AI-driven test-feedback-fix loops.30 npm4MIT
- AlicenseNot gradedqualityAmaintenanceA source-aware MCP server that connects AI agents to browser and server runtimes, enabling real-time debugging, monitoring, and automatic fixes via WebSocket or HTTP.2MIT
- AlicenseNot gradedqualityDmaintenanceAn MCP server that provides CVE-driven security prompts for code review, enabling coding agents to get actionable security checks based on real vulnerabilities.MIT