Install
$ agentstack add mcp-zephexmcp-zephex-tools Open-source listing, not yet scanned by AgentStack. Follow the source repository for install instructions.
Security review
⚠ Flagged1 finding(s); flagged for manual review. · v0.1.0 How review works →
- • Prompt-injection patterns
- • Secret / credential exfiltration
- • Dangerous shell & filesystem operations
- • Untrusted network calls
- • Known-malicious package signatures
- high Pipes remote content directly into a shell (remote code execution).
What it can access
- ● Network access Used
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
Zephex Tools
The ten codebase intelligence tools — full reference for agents and developers.
Same tools, same API key, same credits, three surfaces:
| Surface | Repo | Best when | |---------|------|-----------| | Editor MCP | zephex-MCPs | Cursor / Claude Code / Codex / OpenCode | | Local CLI | zephex-cli | Shell, monorepos, deep --json, CI | | Web terminal | zephex-web-terminal | Browser demo, no install, public GitHub |
This repo is the deep tool catalog. Product overview lives on zephex-MCPs. Endpoint: https://zephex.dev/mcp. Free tier: 555 req/mo.
get_project_context → find_code → read_code → [edit] → check_test
↘ explain_architecture when the change spans modules
check_package before any new dependency
project_memory when you learn something non-obvious
keep_thinking only if still stuck after real tool use
audit_headers only on a user-supplied live URL
Zephex_dev_info for generic playbooks (not their private code)
Contents
- [Canonical names](#canonical-names-only-these-ten)
- [Project source](#project-source-almost-every-code-tool)
- [Default agent chain](#default-agent-chain)
- [Tool reference](#tool-reference) — all ten with examples
- [CLI map](#cli-map-mode-2)
- [Workflows](#concrete-workflows)
- [Connect](#connect-once-all-surfaces)
- [For AI research](#for-ai-research--scrapers)
On-site guides (same tools, richer UI docs): zephex.dev/docs/tools
Canonical names (only these ten)
| # | Tool | One-liner | |---|------|-----------| | 1 | get_project_context | Stack, scripts, auth, deploy — one topic per call | | 2 | find_code | Ranked search: snippet · symbol · concept · everywhere | | 3 | read_code | AST symbol / file batch / outline / scan / smell | | 4 | explain_architecture | Wiring map, layers, edges; deep + Mermaid | | 5 | check_package | Registry safety · upgrade notes · CVEs (12 ecosystems) | | 6 | check_test | Test Pulse — run, failures, fix_prompt, missing | | 7 | audit_headers | Live HTTPS grade (user-supplied URL only) | | 8 | project_memory | remember / recall / list / forget across sessions | | 9 | keep_thinking | Structured multi-step reasoning + loop detection | | 10 | Zephex_dev_info | Expert playbooks (Stripe, RLS, …) — not private code |
Dead names (removed — never document as live)
| Wrong | Right | |-------|--------| | scope_task | find_code + architecture / CLI deep | | inspect_url | audit_headers | | audit_package | check_package with task | | bare thinking | keep_thinking |
Server may still alias some old names; do not teach them as product names.
Project source (almost every code tool)
Pass one of:
| Mode | Example | |------|---------| | Local disk (stdio / CLI) | path: "/absolute/to/project" | | Public GitHub | path: "github:owner/repo" or full HTTPS URL | | No disk (hosted only) | inline_files: { "src/a.ts": "" } |
Hosted MCP often blocks absolute laptop paths — use GitHub URL or inline_files.
Default agent chain
1. get_project_context topic=identity (then run, then auth if needed)
2. find_code when location unknown
3. read_code when path or symbol known
4. [agent edits in the editor]
5. check_test task=run → then free failures / fix_prompt on session_id
| Situation | Tool | |-----------|------| | Cross-cutting feature (auth, billing, monorepo) | explain_architecture | | Before npm install / version bump | check_package | | User pasted a production URL | audit_headers | | “Remember we use X” / “what did we decide” | project_memory | | 2+ failed debug attempts, high blast radius | keep_thinking | | Generic “how do Stripe webhooks work” | Zephex_dev_info |
Tool reference
Credits below are approximate (verify in dashboard / response meta). Free pool is shared across MCP + CLI + web terminal.
1. get_project_context — Project Context
~7 credits · start here on any unknown repo
Docs: get-project-context
Reads manifests (package.json, pyproject.toml, Cargo.toml, go.mod, …) and returns one focused topic per call. Not for symbol search or file bodies.
| | | |--|--| | Topics | identity · framework · backend · frontend · database · auth · deploy · run · structure · integrations · security | | Input | path or inline_files; topic; detail_level brief\|standard\|full; optional force, include_structure, focus_on | | Output | topic, focus, summary, data.key_paths, hint, related_topics, next_calls | | When | First touch before guessing install/test commands | | CLI | zephex overview · zephex get-context / context / stack · zephex deep (composite) |
Example (MCP)
{
"path": "github:owner/repo",
"topic": "identity",
"detail_level": "standard"
}
Example (CLI)
cd your-app
zephex overview
zephex get-context --topic run
zephex deep "add rate limiting" --json # agent packet schema_version: 1
Do not use for find/read/test — those are other tools.
2. find_code — Code Search
~5 credits · location unknown
Docs: find-code
Ranked search inside the project. Pick intent, not a pile of knobs.
| Intent | Use for | |--------|---------| | snippet | Pasted line of code | | symbol | Known name (function, class, const) | | concept | Topic + optional also_try synonyms | | everywhere | Rename map (whole_word recommended) |
| | | |--|--| | Input | query (required); path; intent; also_try[]; include code\|docs\|config\|data\|all; file_pattern; whole_word; case_sensitive; response_format | | Output | matches / all_occurrences, files_hit, files_summary, next_calls | | When | Unknown location; rename before edit; multi-keyword hunt | | CLI | zephex find "…" · find-code · defs · rename · paste |
Example (MCP)
{
"path": "github:owner/repo",
"query": "rate limit middleware",
"intent": "concept",
"also_try": ["throttle", "ratelimit"]
}
Example (CLI)
zephex find "auth middleware"
zephex rename OldName # intent everywhere
Then read_code on the top hit from next_calls.
3. read_code — Code Reader
~5 credits · path or symbol known
Docs: read-code
Surgical AST read with a token budget. Prefer after find_code.
| Mode | Does | |------|------| | symbol | Extract by name (default) | | file | Batch 1–20 paths | | outline | Structural TOC — use first on 300+ line files | | scan | Keywords across files[] | | smell | Bug patterns on focus files | | callers / blast_radius / dead_code | Local disk only |
| | | |--|--| | Input | path; mode; target / symbol_id / targets[]; files[]; offset_line / limit_lines; compact; inline_files | | Output | signatures, bodies, usages, hint, next_calls | | When | Known symbol or files from find_code | | CLI | zephex read · outline · symbol · summarize · files |
Example (MCP)
{
"path": "github:owner/repo",
"mode": "outline",
"files": ["src/middleware/auth.ts"]
}
Tips: files under ~50 lines → native editor Read is fine. Batch file mode returns every path in one call. Pass max_results: 10 when batching many targets.
4. explain_architecture — Architecture Analyzer
~7 credits · cross-cutting change
Docs: explain-architecture
How modules wire together — not full file bodies. Seeds via concept search + import graph.
| | | |--|--| | Input | path; concern (any label) or seed_files[]; focus api\|auth\|integrations\|database\|security\|full; mode overview\|deep\|audit; subpath for monorepos | | Output | concern_cluster, layer_map, entry_points, integration_map, auth_flow; deep adds request_flows + Mermaid | | When | “How does auth / billing / this feature span the repo?” | | CLI | zephex architecture · arch [--focus auth] |
Example (MCP)
{
"path": "github:owner/repo",
"concern": "authentication",
"mode": "deep"
}
Example (CLI)
zephex architecture
zephex arch --focus auth
5. check_package — Package Check
~5 credits · before install or bump
Docs: check-package
Live registry across 12 ecosystems. One tool, multiple tasks (not a separate “audit_package”).
| Task | Does | |------|------| | check | Exists / typosquat / CVE gate | | security | CVEs only | | upgrade / migrate | Breaking changes, migration steps, examples from GitHub releases | | debug | Advisories + extra detail |
| | | |--|--| | Input | package (required); task; version / from_version; source github:owner/repo\|local; ecosystem | | Output | exists, CVEs, risksignals, breakingchanges, hint | | When | “Is X safe?” · “How do I upgrade Next?” | | CLI | zephex safe lodash · zephex check-package next --task upgrade --from-version 14 |
Example (MCP)
{
"package": "lodash",
"task": "check"
}
{
"package": "next",
"task": "upgrade",
"from_version": "14.2.0",
"source": "github:owner/repo"
}
If data.exists: false → do not install.
6. check_test — Test Pulse
~1 credit per hosted run; follow-ups free with session_id
Docs: test-pulse
Executes the detected test runner and returns agent-native health — not task scoping.
| Task | Credits | Does | |------|---------|------| | run | ~1 hosted | Run suite, return health + session_id | | detect / missing | 0 | Inventory / test-gap scan | | failures · list · why · fix_prompt · status · … | 0 | After a run, on session_id |
| | | |--|--| | Input | path; task; session_id; optional area / file_filter / diff_base / with_coverage | | Output | health, broken_areas, fix_first, failure_clusters, blind_spots, session_id | | When | After edits, before push, “do tests pass?” | | CLI | zephex test · zephex check test failures · zephex check test fix-prompt --copy |
Example (MCP)
{ "path": "github:owner/repo", "task": "run" }
{ "task": "failures", "session_id": "" }
Example (CLI)
zephex test
zephex check test failures
zephex check test fix-prompt --copy
Hosted: public GitHub or inline_files. Local absolute paths on hosted MCP are blocked; stdio/CLI runs on your machine.
7. audit_headers — Site Audit
~5 credits · user pastes a live URL
Docs: audit-headers
Public URL intelligence on Zephex servers: security grade A–F, health, tech stack, optional secret scan. Blocks localhost / private IPs.
| | | |--|--| | Input | url (required); scan_mode quick\|thorough; scan_depth quick\|deep; probe_engine fetch\|browser; timeout_ms | | Output | plain_summary, grade, site_verdict, trust score, tech, secrets, headers | | When | Post-deploy, “is it up”, security headers | | CLI | zephex check url https://example.com · site |
Example (MCP)
{
"url": "https://example.com",
"scan_mode": "quick",
"probe_engine": "fetch"
}
Not for repo code (find_code) or tests (check_test).
8. project_memory — Project Memory
~1–4 credits · across sessions
Docs: project-memory
Persist decisions, gotchas, goals, conventions per project path.
| Action | Does | |--------|------| | remember | Save (title ≤80, content ≤500) | | recall | Keyword search | | list | Titles | | forget | By id |
| | | |--|--| | Types | decision · gotcha · goal · preference · area_fact · convention | | When | User says remember/recall; after non-obvious fix | | CLI | zephex remember · recall · memory list |
Example (MCP)
{
"action": "remember",
"path": "github:owner/repo",
"title": "Auth uses JWT in cookies",
"content": "Access token 15m, refresh 7d, SameSite=Strict",
"type": "decision"
}
Scoped by project — not a global dump. Paraphrase facts; do not paste whole files. Not a substitute for get_project_context.
9. keep_thinking — Keep Thinking
~1–4 credits · stuck / high blast radius
Docs: keep-thinking
Multi-call reasoning scaffold with loop detection and risk gates. Not a replacement for search/read.
| | | |--|--| | Required | thought, thoughtNumber, totalThoughts, nextThoughtNeeded, confidence, thoughtType | | Useful | goalAnchor (after thought 2), lastActions[], sessionId | | When | Auth/billing/schema edits; 2+ failed debug attempts; 3+ file tasks | | CLI | zephex think "…" · reason (one-shot; multi-turn is editor-native) |
Pass lastActions or loop detection is weak. Close with nextThoughtNeeded: false. Cap ~10 thoughts/session.
10. Zephex_dev_info — Expert playbooks
~2–5 credits · generic patterns only
Docs: zephex-dev-info
Internal knowledge base: Stripe webhooks, Supabase RLS, CSP, Next.js auth, mobile signing, etc. Not the user’s private repo.
| | | |--|--| | Flow | operation: search → then operation: get with returned slug | | Categories | databases · security · frontend · backend · auth · mobile · android · payments | | When | After repo tools if you still need a standard playbook | | CLI | zephex docs "…" · ask "…" |
Complements get_project_context. Not for package registry intel (check_package).
CLI map (Mode 2)
Same backend. Type these in a real shell after cd into the project (or --cwd).
| You type | MCP tool | |----------|----------| | overview / deep / get-context | get_project_context (+ composites) | | find / find-code / rename | find_code | | read / outline / symbol / files | read_code | | architecture / arch | explain_architecture | | safe / check-package | check_package | | test / check test … | check_test | | check url / site | audit_headers | | remember / recall | project_memory | | think / reason | keep_thinking | | docs / ask | Zephex_dev_info | | structure --agent | local layout (0 credits) |
Install CLI
curl -fsSL https://zephex.dev/cli/install.sh | bash
# or
npx zephex setup
cd your-project
zephex login
zephex deep --json # orientation packet for agents
zephex overview
zephex find "auth"
zephex test && zephex check test failures
zephex safe lodash
Full CLI surface: zephex-cli · docs/cli-commands
Web terminal (browser, allowlisted commands): zephex-web-terminal
Concrete workflows
Onboard a repo
1. get_project_context topic=identity
2. get_project_context topic=run
3. get_project_context topic=auth (if relevant)
4. explain_architecture mode=overview
5. find_code for entry + auth middleware
6. Summarize stack + real paths — do not invent structure
Safe package
…
Source & license
This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: zephexMCP
- Source: zephexMCP/zephex-tools
- License: MIT
- Homepage: https://zephex.dev/docs/tools
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.