Get started

Coding agents

Set up Parlo with Claude Code, Codex, Cursor, and other coding agents.

Parlo works with any coding agent that can run shell commands. Jump to your agent:

There are three ways to connect Parlo to your agent:

1MCP: our hosted server at https://parlo.run/mcp. Best for tool-call clients. See the MCP guide, which covers Cursor, Claude Code, Codex, and OpenCode.
2CLI and mount: the agent runs parlo commands against the ~/parlo mount. Best for shell agents. Set it up below with the skill (recommended) or a manual instruction block.
3Paste a URL: if you already have an OpenAPI, GraphQL, or MCP server URL, paste it on the connections page or run parlo connect --spec <url> / parlo connect --mcp <url>. No wizard needed.

Which should you pick? If your agent has a shell (Claude Code, Codex), the CLI is usually cheaper and more reliable: it already knows shell commands and skips loading a tool schema into context. Reach for MCP when your client prefers tool calls or cannot run a shell. They share the same connections and drives, so you can use both.

Option 1: Install the skill

Parlo ships a skill file that teaches your agent its full command surface. Drop it into your agent's skills folder and it loads automatically on the next session.

Claude Code

bash

2 lines

1mkdir -p ~/.claude/skills/parlo2curl -fsSL https://parlo.run/skill/parlo/SKILL.md -o ~/.claude/skills/parlo/SKILL.md

Codex

bash

2 lines

1mkdir -p ~/.agents/skills/parlo2curl -fsSL https://parlo.run/skill/parlo/SKILL.md -o ~/.agents/skills/parlo/SKILL.md

OpenCode

bash

2 lines

1mkdir -p ~/.config/opencode/skills/parlo2curl -fsSL https://parlo.run/skill/parlo/SKILL.md -o ~/.config/opencode/skills/parlo/SKILL.md

Start a new session and your agent picks up Parlo automatically. To scope it to one project instead, swap the ~/.claude or ~/.agents prefix for a local .claude or .agents folder in your repo.

Option 2: Manual setup

If you prefer not to install the skill, or your agent loads a single config file, add the instruction block below to that file.

Install Parlo

Install with npm or the install script:

bash

3 lines

1npm install -g parlofs@latest2# or3curl -fsSL https://parlo.run/install | sh

bash

3 lines

1parlo setup             # browser sign-in2parlo connect github    # connect a provider3parlo connections       # verify what's connected

The instruction block

Copy this into your agent's config file. The sections below explain which file to use for each agent.

markdown

Parlo

Parlo exposes cloud drives, documented APIs, GitHub repos, and remote code search as files. Reach for it whenever a task touches connected services (GitHub, Linear, Stripe, search/web APIs), cloud files, or large-repo search.

Discover what's available

bash

3 lines

1parlo help                          # full command list and flags2parlo connections                   # what's connected3parlo cat <connection>/README.md    # how each connection is laid out

Run parlo help and read each connection's README.md before assuming what exists. They document the real commands, paths, and conventions.

Core commands

bash

9 lines

1# API + drive files2parlo ls <connection>/<path>                       # list resources3parlo cat <connection>/<path>                       # read a resource as markdown4parlo write <connection>/<path> --from <file|->     # create or update5parlo rm <connection>/<path>                        # delete (gated, opt-in per client)6 7# Remote code search (runs server-side, no clone)8parlo rg <pattern> <owner/repo> --path src          # raw ripgrep9parlo grep <pattern> <owner/repo>                   # formatted results

If ~/parlo is mounted, use normal shell tools instead:

bash

3 lines

1ls ~/parlo/github/issues/2grep -R "billing" ~/parlo/github/issues3cat ~/parlo/github/issues/42.md

File conventions

bash

4 lines

1README.md                usage guide for this level (read first)2recent.jsonl             latest items, fast snapshot3all.jsonl                full dataset export (first read may take ~30s)4q/  results/  details/   queries, compact indexes, full detail

*.md is best for reading, *.jsonl pipes into jq, pandas, or DuckDB.

Rules

Run parlo help and read README.md files to discover features. Do not guess.
Use grep and compact indexes before reading large detail files.
Never print, log, or commit API tokens or files under ~/.parlo.

Claude Code

Add the instruction block above to CLAUDE.md at the root of your project. Create the file if it does not exist.

Claude Code automatically loads CLAUDE.md at the start of every session. It has a Bash tool, so it can run parlo commands directly and also use normal shell commands against the ~/parlo mount.

Codex

Add the instruction block above to AGENTS.md at the root of your project. Create the file if it does not exist.

Codex auto-injects the root AGENTS.md into the developer message at the start of every task. You can also scope instructions to a subdirectory by placing an AGENTS.md inside it. Codex applies the nearest ancestor file.

Network access note: parlo cat makes HTTP calls to the Parlo API. In Codex's read-only or workspace-write sandbox modes, the first call may require user approval. In danger-full-access mode, everything works without prompts.

Cursor

Create .cursor/rules/parlo.mdc. Start with this frontmatter, then paste the instruction block below it:

text

5 lines

1---2description: Use Parlo for connected cloud services and APIs3globs:4alwaysApply: true5---

The alwaysApply: true field tells Cursor to include this rule in every conversation, regardless of which files are open.

Other agents

Any agent with shell access can use Parlo. Put the instruction block in whatever config file the agent reads:

Agent	Config file
Windsurf	`.windsurfrules`
Aider	`.aider/conventions.md`
OpenCode	`AGENTS.md`
Custom agents	Whatever system prompt or config the agent loads

The pattern is always the same: paste the instruction block, and the agent will call parlo commands or read from ~/parlo as needed.

CI and sandboxed agents

Agents running in containers or CI environments often cannot mount FUSE filesystems. Use direct ops mode instead:

bash

8 lines

1# Add a connection headlessly using a provider token2parlo connect github --headless --token "$GH_PAT"3 4# Then use direct ops5parlo connections6parlo ls github/issues/7parlo cat github/issues/42.md8parlo write main/reports/summary.md --from ./summary.md

If no Parlo account exists in the environment yet, run parlo agent-signup for proof-of-work signup without a browser. For human-gated signups (Turnstile, email verification), ask a team member to run parlo setup first and share the session.

Verify it works

Ask the agent:

> List my Parlo connections.

Or give it a real task:

> Use Parlo to find GitHub issues about billing and write a short summary.

If the agent runs parlo connections and lists your connected services, everything is working.

← Previous

Quickstart

MCP