Fastest: npx -y zephex setup --crush (browser sign-in, writes crush.json). Or paste the mcp.zephex block below. Zephex uses https://zephex.dev/mcp with Bearer auth in the file — internet required.
Official Crush MCP documentation: Crush documentation
Without MCP, your agent guesses project layout and misses supply-chain risk. Zephex connects one hosted endpoint so every session gets the same ten tools — no per-machine npm installs, no version drift across the team.
HTTP and stdio setups both need a key from your Zephex dashboard. OAuth-only flows (ChatGPT, Claude.ai web) sign you in in the browser instead — skip this section for those.
Matches the published CLI (mcp-proxy/src/commands/setup.ts). One command signs you in, writes the correct transport, and verifies 10 tools.
Account teardown: logout vs disconnect · Connect MCP walkthrough
Search docs for “where is my MCP file” — the answer is always: run list first, then open the path it prints.
Documented paths: crush.json (project) or ~/.config/crush/crush.json (global)
Crush crush.json uses mcp.zephex with type http — matches Charm schema.
Runs the same code as mcp-proxy/src/commands/setup.ts — OAuth in browser, creates a CLI key, writes your editor config, verifies tools.
npx -y zephex setup --crushSkip browser OAuth — paste a key from zephex.dev/dashboard/keys. Must start with mcp_prod_, mcp_dev_, or mcp_sk_.
npx -y zephex setup --crush --api-key mcp_prod_your-key-hereRuns the same code as mcp-proxy/src/commands/setup.ts — OAuth in browser, creates a CLI key, writes your editor config, verifies tools.
npx -y zephex setup --crush --projectmcp.zephex HTTP block — merge into crush.json: Transport: http (see crush.json (project) or ~/.config/crush/crush.json (global)).
{ "$schema": "https://charm.land/crush.json", "mcp": { "zephex": { "type": "http", "url": "https://zephex.dev/mcp", "headers": { "Authorization": "Bearer mcp_sk_your_key_here" } } }}After CLI install, fully restart the app if tools do not appear. Manual JSON/TOML blocks below are equivalent — use them when CLI commands are unavailable.
Full comparison: HTTP vs stdio · npx zephex reference
Crush uses a top-level mcp key — not Cursor’s mcpServers wrapper. Wrong shape is the most common mistake.
Replace mcp_sk_your_key_here with your key from Dashboard → API Keys. Copy the full key once at creation — paste into Authorization: Bearer … for HTTP configs, or into ZEPHEX_API_KEY for stdio/npx configs.
Type must be "http". Do not use command/npx for Zephex — that runs a local process instead of the hosted endpoint.
{ "$schema": "https://charm.land/crush.json", "mcp": { "zephex": { "type": "http", "url": "https://zephex.dev/mcp", "headers": { "Authorization": "Bearer mcp_sk_your_key_here" } } }}Tip
Launch Crush from the project directory when you want project crush.json to apply; global config applies when no project file exists.
Run these in a terminal when the editor UI is unclear — catches stale npm, wrong transport, and project shadows.
npx -y zephex@latest listnpx -y zephex@latest doctornpx -y zephex@latest repair# Fully quit the editor (Cmd+Q / Alt+F4), reopen, start a new agent sessionAfter saving your config, confirm Zephex is connected before you rely on it in real work.
Questions people ask when Crush does not show Zephex tools — indexed for docs search.
How do I connect Zephex to crush?
Fastest: npx -y zephex setup --crush (browser sign-in, writes config, verifies 10 tools). Or paste manual config on this page, save, fully quit the app, reopen.
Where did setup save my crush MCP config?
Run npx -y zephex@latest list — it prints every config path on this machine that references Zephex.
crush works in terminal but not in the editor
Confirm Bearer header and url https://zephex.dev/mcp. Revoke old keys and reconnect if you see 401.
How do I disconnect Zephex from crush?
mcpcli disconnect --crush — revokes key and strips config.
You do not call tools yourself — ask your agent in plain language. Try these once Zephex is connected:
“In Crush TUI, check_package on a Go module before go get in this Charm workspace.”
Terminal agent gets registry intel over HTTP MCP.
“get_project_context on the repo where crush.json lives.”
Stack snapshot for Bubble Tea / Go layouts.
“read_code the Update function in our view model — compact mode on.”
Small focused reads fit narrow terminal panes.
“find_code tea.KeyMsg handlers in this codebase.”
Idiomatic search for Charm ecosystem patterns.
“check_test: add config file hot-reload to our CLI.”
Minimal touch list before Crush edits multiple packages.
“audit_headers on our project's public demo URL.”
Quick security sanity check from the same session.
You do not pick tools from a menu — ask your agent in normal sentences. Half of the tools only need a package name or URL; the other half need to know which codebase you mean (your Mac/Windows project folder or a GitHub repo).
These situations usually mean the setup cannot work until you fix the underlying issue:
Wrong JSON shape
Use mcp.zephex at top level per crush.json schema — not mcpServers.
Config not loaded
Check project crush.json vs ~/.config/crush/crush.json. Restart Crush from the intended cwd.
401 Unauthorized
"Authorization": "Bearer mcp_sk_..." — quotes around the full header value in JSON.
type http not recognized
Update Crush to a build that supports HTTP MCP per Charm docs.
find_code returns nothing
Name the repo: github:owner/repo or absolute path on Mac/Windows in the same message.
Crush http MCP block unlocks ten developer tools in the TUI agent:
get_project_context
Reads your project structure, dependencies, scripts, env vars, and framework markers in one call. Replaces manually opening package.json, tsconfig, and multiple config files at the start of every session.
read_code
AST-based code extraction: pass a symbol name and get the implementation without reading entire files. Supports symbol lookup, batched file reads, and structural outlines for large files.
find_code
Ranked search across the repo for definitions, usages, and patterns. Faster than blind grep when the agent does not know where a symbol lives.
check_package
Live registry lookup for npm, PyPI, Cargo, and Go modules. Surfaces typosquat risk, maintainer changes, and suspicious version jumps before you run install.
explain_architecture
Generates Mermaid diagrams for auth flows, service boundaries, and module dependencies so the agent reasons about structure instead of guessing.
check_test
Turns a task description into the smallest file set to read or edit, with risk ratings and caller impact notes.
audit_headers
Grades a deployed URL for CSP, HSTS, TLS, cookies, and redirects. Returns fix snippets for common hosts (Vercel, Cloudflare, Nginx).
keep_thinking
Structured multi-step debugging: tracks hypotheses and conclusions so long investigations do not loop.
Zephex_dev_info
Expert patterns for authentication, databases, frontend frameworks, deployment, and mobile stacks when the agent needs vetted guidance.
project_memory
Persists decisions, gotchas, and conventions per project in ~/.zephex/memory (SQLite FTS5). recall before unfamiliar areas; remember after discoveries. Local stdio only on npx zephex.