# Warp

The Agentic Development Environment — a terminal and agent platform with local and cloud agents (Oz), MCP, rules, and skills.

Schema: 1.0 · Data: ee5de863c30bdbe4fe7dd3e07c7e22ea9bf4ab2248dbc7c35cdbf60bc28be1db

## Configuration

### MCP

Status: supported

MCP servers configured in Settings > Agents > MCP servers, in Warp Drive, or via file-based .warp/.mcp.json config. Warp also picks up MCP config from Claude Code, Codex, and other agent providers.

- global: `~/.warp/.mcp.json` (json) — key: mcpServers — Shared by Stable and Preview channels. Pasted multi-server JSON also uses the mcpServers key.
- project: `.warp/.mcp.json` (json) — key: mcpServers — Project-scoped; requires explicit approval and never auto-spawns.
- `command`: string (required) — Executable to launch for a CLI (stdio) server, e.g. npx.
- `args`: string[] (required) — Command-line arguments passed to the command.
- `env`: record<string, string> — Environment variables for the server (e.g. API tokens).
- `working_directory`: string — Working directory where the command runs; used to resolve relative paths.
- `url`: string (required) — HTTP endpoint for a Streamable HTTP or SSE server.
- `headers`: record<string, string> — HTTP headers for remote servers (e.g. Authorization bearer token).
- transports: stdio, sse, http
- auth: oauth, headers, env
- Tools: supported
- Resources: supported — Running servers list their available tools and resources on the MCP servers page.
- Global Warp-managed servers auto-spawn by default; global servers from third-party providers (Claude Code ~/.claude.json, Codex ~/.codex/config.toml, other agents ~/.agents/.mcp.json) require the Auto-spawn servers from third-party agents toggle.
- Project-scoped servers from any provider never auto-spawn and must be approved/toggled on per session.
- The bundled /agent-add-mcp skill lets the agent write file-based server definitions to ~/.warp/.mcp.json (global) or {repo_root}/.warp/.mcp.json (project) with your approval.
- Warp prevents edits to MCP config files unless you explicitly approve the change.
- Servers can be shared with teammates via Warp Drive; env secrets are scrubbed and replaced with variables on share.
- Set working_directory explicitly when command/args include relative paths.

Multiple servers (paste or .warp/.mcp.json)
```json
{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/allowed/files"]
    },
    "externalDocs": {
      "url": "http://localhost:4000/mcp/stream",
      "headers": { "my-header": "my-header-value" }
    }
  }
}

```

Source: https://docs.warp.dev/agent-platform/capabilities/mcp/
Checked: 2026-07-12

### Skills

Status: supported

SKILL.md skills (Agent Skills format) discovered from .agents/skills/ and many compatible directories, invocable by the agent or directly via /{skill-name} with argument placeholders.

- project: `.agents/skills/<name>/SKILL.md` (markdown) — Recommended project location; subdirectory skills are discovered when working within that subdirectory.
- global: `~/.agents/skills/<name>/SKILL.md` (markdown) — Recommended global location, available across all projects.
- project: `.warp/skills/<name>/SKILL.md` (markdown) — Warp-specific alternative; .claude/skills/, .codex/skills/, .cursor/skills/ and others are also scanned.
- global: `~/.warp/skills/<name>/SKILL.md` (markdown) — Warp-specific global directory; also where Warp's bundled skills live.
- `name`: string (required) — Unique identifier for the skill (typically kebab-case).
- `description`: string (required) — What the skill does and when to use it; the agent uses this to decide when to invoke the skill.
- invocation: /skill-name, automatic
- Warp scans many skill directory names at both project and home scope: .agents/skills/ (recommended), .warp/skills/, .claude/skills/, .codex/skills/, .cursor/skills/, .gemini/skills/, .copilot/skills/, .factory/skills/, .github/skills/, .opencode/skills/.
- Skill discovery is based on the cwd; for git repos Warp includes all skills from the current directory up through the repository root.
- Argument placeholders: $ARGUMENTS (full raw string), $ARGUMENTS[N] (Nth whitespace-separated arg, 0-indexed), $N shorthand. Extra text with no placeholders is passed as a separate user message.
- Skills can include supporting files (scripts, templates, configs) referenced by relative path from SKILL.md.
- Edit skills with the /open-skill slash command; bundled skills live in ~/.warp/skills/.
- Warp maintains ready-to-use skills in the warpdotdev/oz-skills repository; skills can also drive cloud agents (Skills as Agents).

Parameterized skill
```markdown
---
name: explain-topic
description: Explain a topic for a specific audience in a given tone
---

# Explain Topic

Explain $0 for an audience of $1 professionals.
Use a $2 tone.
Full request: $ARGUMENTS

```

Source: https://docs.warp.dev/agent-platform/capabilities/skills/
Checked: 2026-07-12

### Rules

Status: supported

Project Rules in AGENTS.md (or legacy WARP.md) files at repo root and subdirectories, plus cloud-managed Global Rules in Warp Drive.

- project: `AGENTS.md` (markdown) — Repo root and subdirectories; closest file takes precedence.
- project: `WARP.md` (markdown) — Legacy name, still fully supported; wins over AGENTS.md in the same directory.
- AGENTS.md is the default project rules file; existing WARP.md files remain fully supported and take priority if both exist in the same directory.
- The filename must be all caps (AGENTS.md, not agents.md).
- Warp automatically applies the rules file in the repo root and the current directory, and makes a best-effort attempt to include a subdirectory's rules file when editing files there.
- Precedence: current subdirectory rules > root directory rules > Global Rules.
- /init generates an AGENTS.md or links an existing rules file (CLAUDE.md, .cursorrules, AGENT.md, GEMINI.md, .clinerules, .windsurfrules, .github/copilot-instructions.md); /add-rule creates a Global Rule; /open-project-rules opens project rules in Warp's editor.
- Global Rules are managed in Warp Drive (Personal > Rules), not as local files; rules usage can be toggled with the agents.knowledge.rules_enabled setting.

Generate project rules
```text
/init

```

Source: https://docs.warp.dev/agent-platform/capabilities/rules/
Checked: 2026-07-12

### Hooks

Status: unsupported

Warp has no lifecycle hook mechanism for its agents.

- No hooks surface exists in the official docs (docs.warp.dev). Agent lifecycle hooks are open feature requests tracked in warpdotdev/Warp issues #7834, #8931, and #6857.
Checked: 2026-07-12

### Commands

Status: supported

Slash commands in Agent Mode combine built-in static commands (/init, /agent, /model, /add-mcp, ...) with user-defined Agent Prompts saved in Warp Drive and skills invoked as /{skill-name}.

- `name`: string (required) — Prompt name shown in the slash menu and used for search.
- `description`: string — Meaningful description, indexed for search.
- `arguments`: array — Prompt arguments created via {{argument}} placeholders; each has a type (text or enum), optional description, and default value.
- invocation: /command-name, /skill-name, command palette
- Built-in static commands include /agent, /new, /init, /index, /model, /add-mcp, /add-prompt, /add-rule, /prompts, /queue, /create-environment, and more.
- Custom reusable prompts are Warp Drive Prompts — parameterized natural-language queries stored in Warp Drive (cloud objects, not local files) and runnable from the slash menu, Command Palette, or Warp Drive.
- Prompt arguments use {{argument}} placeholders, typed text or enum, with optional descriptions and default values; SHIFT-TAB cycles through arguments.
- Skills are also user-invocable as slash commands with argument support (see Skills).
- Slash command availability in the terminal input is controlled by the terminal.input.enable_slash_commands_in_terminal setting.

Parameterized Warp Drive prompt
```text
Summarize the git log since {{ref}} and highlight breaking changes for {{audience}}.

```

Source: https://docs.warp.dev/agent-platform/capabilities/slash-commands/
Checked: 2026-07-12

### Settings

Status: supported

TOML settings file with hot reload; the [agents] sections control model, knowledge (rules), MCP auto-spawn, and permission profiles for the agent.

- global: `~/.warp/settings.toml` (toml) — macOS Stable path; Linux uses ~/.config/warp-terminal/settings.toml. There is no project-scoped settings file.
- `agents.model`: string — Default model for the Warp Agent in the terminal. Default "auto".
- `agents.knowledge.rules_enabled`: boolean — Whether the agent uses your saved rules during requests. Default true.
- `agents.mcp_servers.file_based_mcp_enabled`: boolean — Auto-spawn global file-based MCP servers from third-party agents. Default false.
- `agents.profiles.agent_mode_coding_permissions`: string — File read permission level for the agent. Values: always_ask_before_reading, always_allow_reading, allow_reading_specific_files
- `agents.profiles.agent_mode_execute_readonly_commands`: boolean — Whether the agent can auto-execute read-only commands without asking. Default false.
- `agents.profiles.agent_mode_command_execution_allowlist`: string[] — Regex patterns for commands the agent can execute without explicit permission.
- `agents.profiles.agent_mode_command_execution_denylist`: string[] — Regex patterns for commands the agent must always ask before executing.
- `agents.warp_agent.is_any_ai_enabled`: boolean — Master switch for all AI features. Default true.
- `agents.warp_agent.other.agent_attribution_enabled`: boolean — Add an attribution co-author line to commits and PRs the agent creates. Default true.
- `agents.warp_agent.other.default_prompt_submission_mode`: string — Behavior when submitting a prompt while the agent is still responding. Default interrupt. Values: interrupt, queue
- `agents.cloud_conversation_storage_enabled`: boolean — Whether conversations are stored in the cloud. Default true.
- `general.default_session_mode`: string — Default mode for new terminal sessions. Default terminal. Values: terminal, agent, cloud_agent, tab_config, docker_sandbox
- `code.indexing.agent_mode_codebase_context`: boolean — Whether Codebase Context is provided to the agent. Default true.
- File location depends on platform/channel: macOS Stable ~/.warp/settings.toml, Linux ~/.config/warp-terminal/settings.toml, Windows %LOCALAPPDATA%\warp\Warp\config\settings.toml.
- Settings are grouped into TOML sections; only include settings you want to change. A settings_schema.json ships with the app.
- Agent-relevant sections: [agents], [agents.knowledge], [agents.mcp_servers], [agents.profiles], [agents.warp_agent.*], [agents.third_party], [agents.voice].
- Settings can sync across devices via account.is_settings_sync_enabled.

Agent permissions in settings.toml
```toml
[agents]
model = "auto"

[agents.profiles]
agent_mode_execute_readonly_commands = true
agent_mode_command_execution_allowlist = ["ls(\\s.*)?", "cat(\\s.*)?"]

[agents.mcp_servers]
file_based_mcp_enabled = true

```

Source: https://docs.warp.dev/terminal/settings/all-settings/
Checked: 2026-07-12

## Search, browser & identification

Not researched. Unknown does not mean unsupported.