1. A Pica account — Sign up free at https://app.picaos.com
2. Connected platforms — In the Pica dashboard, connect the services you want OpenClaw to access (Gmail, Slack, HubSpot, Notion, Shopify, etc.)

• Pica Dashboard — https://app.picaos.com (manage connections and API keys)
• Pica MCP Server — https://github.com/picahq/mcp (the MCP server this skill uses under the hood)
• Pica Docs — https://docs.picaos.com
• mcporter — https://github.com/steipete/mcporter (the MCP client CLI that bridges OpenClaw to Pica)

Critical: OpenClaw does NOT support native MCP server configuration in

openclaw.json

. Keys like

mcp

,

agents.defaults.mcp

, or

mcp.servers

are rejected by OpenClaw's config validator and will crash the process with

Unrecognized key

errors. Instead, Pica integration works through skills + mcporter CLI — the agent calls mcporter as a shell command via its exec tool.

tools/package.json

{
  "name": "pica-bridge",
  "version": "2.0.0",
  "type": "module",
  "dependencies": {
    "mcporter": "^0.7.3",
    "@picahq/mcp": "^2.0.4"
  }
}

tools/mcporter.json

{
  "mcpServers": {
    "pica": {
      "command": "node",
      "args": ["/home/node/tools/node_modules/@picahq/mcp/build/index.js"],
      "env": {
        "PICA_SECRET": "$env:PICA_SECRET"
      }
    }
  }
}

Important: Use

node

with the absolute path to the installed module, not

npx

. Using

npx -y @picahq/mcp

causes slow cold-start downloads inside containers. The

$env:PICA_SECRET

syntax is mcporter's environment variable substitution — it reads from the container's environment at runtime.

openclaw.json

skills.load.extraDirs

mcp

{
  "agents": {
    "defaults": {
      "workspace": "/home/node/workspace"
    }
  },
  "skills": {
    "load": {
      "extraDirs": ["/home/node/skills"]
    }
  }
}

/home/node/skills/openclaw-integrations/SKILL.md

1. Do NOT add

   mcp

   to

   openclaw.json

   — OpenClaw rejects it at any level (

   root.mcp

   ,

   agents.defaults.mcp

   ). The process exits with code 1 and the container restart-loops.
2. Do NOT use

   npx

   in mcporter.json — It downloads packages on every call inside containers. Use

   node

   with the absolute path to the pre-installed module.
3. Do NOT install mcporter globally only — The Pica MCP module needs to be locally installed so mcporter can find it at a known absolute path. Symlink the binary to PATH for convenience.
4. Skills directory must exist before OpenClaw starts — If

   extraDirs

   points to a missing path, OpenClaw may ignore it silently.

• Gmail — Read emails, send messages, manage drafts, organize labels
• Slack — Post messages, read channels, manage users
• HubSpot — List contacts, create deals, manage companies
• Notion — Create pages, query databases, update content
• Shopify — Manage orders, products, customers, inventory
• Linear — Create issues, update projects, track progress
• Google Calendar — List events, create meetings, manage schedules
• GitHub — Manage repos, issues, pull requests
• And 190+ more platforms

mcporter call pica.list_pica_integrations --args '{}' --config /home/node/tools/config/mcporter.json

{
  "connections": [
    { "platform": "gmail", "key": "test::gmail::default::abc123" },
    { "platform": "slack", "key": "test::slack::default::def456" }
  ],
  "available": [
    { "platform": "notion", "name": "Notion", "category": "Productivity" }
  ],
  "summary": { "connectedCount": 10, "availableCount": 205 }
}

• connections[].key

  is the

  connectionKey

  — you'll need it for step 4

• connections[].platform

  is the kebab-case platform name for steps 2-4

• available

  lists platforms that CAN be connected but aren't yet
• If the user's desired platform is in

  available

  (not

  connections

  ), tell them to connect it at https://app.picaos.com

mcporter call pica.search_pica_platform_actions --args '{"platform":"gmail","query":"send email"}' --config /home/node/tools/config/mcporter.json

• platform

  (required): kebab-case platform name from step 1

• query

  (required): natural language description of what you want to do

• agentType

  (optional):

  "execute"

  when the user wants to perform an action,

  "knowledge"

  when they want info

{
  "actions": [
    {
      "actionId": "conn_mod_def::GGXAjWkZO8U::uMc1LQIHTTKzeMm3rLL5gQ",
      "title": "Send Email",
      "method": "POST",
      "path": "/gmail/send-email"
    }
  ],
  "metadata": { "platform": "gmail", "query": "send email", "count": 5 }
}

• actions[].actionId

  is what you'll need for steps 3 and 4
• Up to 5 results are returned, ranked by relevance
• For Gmail, prefer actions with paths starting with

  /gmail/

  (these are "enhanced" actions with cleaner request/response formats) over raw API paths like

  users/me/messages/...

mcporter call pica.get_pica_action_knowledge --args '{"actionId":"<id>","platform":"gmail"}' --config /home/node/tools/config/mcporter.json

• actionId

  (required): from step 2

• platform

  (required): kebab-case platform name

mcporter call pica.execute_pica_action --args '{"platform":"gmail","actionId":"conn_mod_def::GGXAjWkZO8U::uMc1LQIHTTKzeMm3rLL5gQ","connectionKey":"test::gmail::default::abc123","data":{"to":"user@example.com","subject":"Hello","body":"Hi there"}}' --config /home/node/tools/config/mcporter.json

• platform

  (required): kebab-case platform name

• actionId

  (required): from step 2

• connectionKey

  (required): from step 1 (

  connections[].key

  )

• data

  (optional): request body object — contents depend on the action (see step 3)

• queryParams

  (optional): query string parameters object

• pathVariables

  (optional): URL path variable substitutions object

• headers

  (optional): additional HTTP headers object

• isFormData

  (optional): boolean, set true for multipart/form-data

• isFormUrlEncoded

  (optional): boolean, set true for URL-encoded form data

--args

data

• Listing integrations — Show a clean list of connected platform names. Don't show connection keys or IDs.
  • Good: "You have Gmail, Slack, and HubSpot connected."
• Search results — Short numbered list with action titles only.
  • Good: "I found these Gmail actions:\n1. Send Email\n2. List Emails\n3. Create Draft"
• Action docs — Summarize required params in plain language.
  • Good: "To send an email, I need: recipient, subject, and body. I can also add CC/BCC."
• Execution results — One sentence describing what happened.
  • Good: "Done — email sent to alice@example.com."
• Errors — Explain in plain language and suggest a fix. No stack traces.

• When a user first asks about integrations, call

  list_pica_integrations

  to show what's available
• If asked "what can you do with X?", search for actions on that platform and summarize the capabilities
• For Gmail, prefer the enhanced actions (paths starting with

  /gmail/

  ) over raw API actions — they have simpler parameters and return decoded, human-readable data
• Always confirm destructive actions (delete, batch operations) with the user before executing
• Never expose connection keys, API secrets, or internal IDs in responses
• If something fails, explain clearly and suggest next steps (e.g., "That platform isn't connected yet — you can add it at https://app.picaos.com")