For agents

Build with Anjal using your favorite AI agent

Anjal is built API-first with a CLI, an MCP server, a typed SDK, and a curated /llms.txt discovery file — so any agent (Claude, Cursor, Windsurf, Cline, ChatGPT, GitHub Copilot agent mode) can self-orient and start integrating without you writing a brief.

CLI

The seamless first-run path. `anjal mcp install` auto-detects every MCP host on the machine and wires it up — no JSON editing, no per-host docs. Cheaper in agent-token cost than MCP for long sessions (Claude Code, CI, terminal-based agents) because the surface is discovered lazily through --help. Includes `listen` for ngrok-free webhook tunneling and `doctor` for a one-shot health check.

Install

npm install -g anjal-cli
anjal login           # opens browser, click Approve, done
anjal mcp install     # auto-wires Claude Desktop / Cursor / Windsurf / Claude Code

anjal-cli on npm

Remote connector (Claude, ChatGPT, Codex)

Add the URL as a custom connector, sign in to Anjal and click Allow. No install and no API key to copy. 47 tools for inboxes, mail, drafts, attachments and calendars, each labelled read-only or destructive. Disconnect any time from Dashboard, API Keys.

Connector URL

https://mcp.anjal.app/mcp

# Claude Code
claude mcp add --transport http anjal https://mcp.anjal.app/mcp

Setup guide

MCP server (local)

Run locally with an API key for Cursor, Windsurf, Cline, or any client that launches a subprocess. Use an inbox (ak_) or workspace (wk_) key to narrow what the agent can reach.

Add to your MCP config

{
  "mcpServers": {
    "anjal": {
      "command": "npx",
      "args": ["-y", "anjal-mcp@latest"],
      "env": { "MYAGENTMAIL_KEY": "tk_..." }
    }
  }
}

anjal-mcp on npm MCP install guide

Claude skill

Drop our SKILL.md into your Claude project. The skill manifest is auto-loaded by Claude when relevant prompts appear, giving the agent our auth + capabilities + recipes for free.

Install

# In your Claude project's .claude/skills/ directory:
curl -o anjal.md https://anjal.app/skills/anjal/SKILL.md

View SKILL.md Anthropic Skills docs

TypeScript SDK

When the agent writes actual code (vs. calling MCP tools or shelling out to the CLI), the typed SDK gives the cleanest shape. Auto-completes, infers types, surfaces docs inline.

Install

npm install anjal

anjal on npm

API reference

Every endpoint, every shape — as HTML for humans and markdown for agents. Each endpoint has a paired markdown alternate at /docs-md/<slug> with schema + code samples in TypeScript, Python, and curl. The OpenAPI spec is the source of truth underneath.

Fetch any endpoint as markdown

curl https://anjal.app/docs-md/domains/listdomains

Browse /docs openapi.yaml

What we built for agents

These are the canonical discovery files. You don't need to read them yourself — your agent does. Listed here so you know what to point an agent at if it needs more than the briefing above.

/llms.txt

Curated discovery index. H2-sectioned link list of every authoritative doc. The default starting point for any agent.

/llms-full.txt

Single-fetch concatenation of SKILL.md + every reference doc (~1,700 lines). Use when context budget allows one round-trip.

/docs

Custom-built API reference. One static page per endpoint with schema tables, response variants, and copy-pasteable TypeScript / Python / curl. Browse the index, then jump straight to the operation you need.

/docs-md/<group>/<operation>

Markdown alternate for any endpoint page. Same content as /docs/<slug>, no chrome — exactly what an agent wants to copy into context. Linked from every endpoint page via the AI actions toolbar.

SKILL.md

Authoritative skill manifest. Auth model + capabilities + Quick Start for agents in the first 80 lines.

references/linkedin.md

LinkedIn surface deep-dive: signals, sessions, multi-session routing, end-to-end recipes.

references/webhooks.md

Webhook event types, HMAC verification, retry schedule, payload schemas.

references/websockets.md

Real-time inbox stream, subscription model, reconnect semantics.

OpenAPI spec

Every endpoint, every shape. Authoritative.

Conventions we follow

Markdown alternates for everything readable. Every blog post has a .md alternate. Every API endpoint has one too at /docs-md/<slug> — schema, responses, and ready-to-run code samples in plain markdown. Skill docs and reference docs are already markdown at their canonical URL.
"Open in Claude/ChatGPT" on every doc page. The AI actions toolbar at the top of every blog post and endpoint page hands the current page (with its markdown URL) directly to your agent in one click — no copy-paste, no context loss.
HTML head exposes the entry point. Every page on this site has <link rel="alternate" type="text/markdown" href="/llms.txt">. HTML-aware crawlers + agents discover the discovery file without needing a hard-coded URL.
All endpoints versioned at /v1/. Breaking changes ship as /v2/ — no silent breakage on /v1.
Webhook payloads HMAC-signed. Verify with the per-signal whsec_... secret and the X-MyAgentMail-Signature header.
Errors are agent-readable. Every error response names the specific field that's wrong, tells the caller what to do next, and never leaks internal mechanics. Agents can self-correct without escalating.

Building a developer product and want to think through your own DX-for-agents strategy? DM me on LinkedIn.