Enable web search with MCP

hrida-agent-sdk can search the web through the Perplexity MCP server. You register the server in ~/.hrida.json, then allow its tools in ~/.hrida-agent-sdk/settings.json.

Run these in the AI Labs terminal

Edit these files on the machine AI Labs runs on, using AI Labs' built-in terminal or file browser. Install hrida-agent-sdk and configure a provider first.

Prerequisites

• A Perplexity API key (pplx-...). Create one in your Perplexity account under Settings → API.
• Node.js and npm, so npx can download and start the MCP server.

Keep your API key private

~/.hrida.json stores the key in plain text. Don't share or commit the file, and don't paste the key into chats or docs.

1. Register the Perplexity server

Open /home/cptr/.hrida.json. It already has an empty "mcpServers": {} entry. Fill it in, replacing pplx-your-key-here with your key:

"mcpServers": {
  "perplexity": {
    "type": "stdio",
    "command": "npx",
    "args": ["-y", "@perplexity-ai/mcp-server"],
    "env": {
      "PERPLEXITY_API_KEY": "pplx-your-key-here"
    }
  }
}

Only change the mcpServers entry. Leave the rest of the file as it is.

2. Allow the Perplexity tools

Open /home/cptr/.hrida-agent-sdk/settings.json and add:

"permissions": {
  "allow": ["mcp__perplexity"]
},

mcp__perplexity allows every tool from the perplexity server, so the agent can search without asking each time.

• If the file already has a "permissions" block, add "mcp__perplexity" to its existing "allow" list instead of adding a second block.
• Keep the file valid JSON: entries are separated by commas, and the last entry before a closing } has no trailing comma.

3. Restart and check

Exit hrida-agent-sdk (/exit) and start it again, then run:

/mcp

perplexity should be listed as connected. You can also check from the terminal:

hrida-agent-sdk mcp list

Then try a search, for example: "Search the web for the latest Node.js LTS version."

Start a new chat in AI Labs to use web search with the hrida-agent-sdk agent.

Other servers you can connect

Any MCP server can be added the same way: add an entry inside mcpServers in ~/.hrida.json, allow it with mcp__<name> in settings.json, then restart.

Web search and browsing

Server	What it adds	API key
brave-search	Web, news, image, and local search through Brave	BRAVE_API_KEY
tavily	Search and page extraction built for AI agents	TAVILY_API_KEY
exa	Semantic web search, company and code search	EXA_API_KEY
firecrawl	Scrape and crawl websites into clean Markdown	FIRECRAWL_API_KEY
fetch	Fetch any URL and convert it to Markdown	None
playwright	Drive a real browser: open pages, click, fill forms, take screenshots	None

"mcpServers": {
  "perplexity": { "...": "as in step 1" },
  "brave-search": {
    "type": "stdio",
    "command": "npx",
    "args": ["-y", "@brave/brave-search-mcp-server"],
    "env": { "BRAVE_API_KEY": "your-brave-key" }
  },
  "tavily": {
    "type": "stdio",
    "command": "npx",
    "args": ["-y", "tavily-mcp"],
    "env": { "TAVILY_API_KEY": "tvly-your-key" }
  },
  "exa": {
    "type": "stdio",
    "command": "npx",
    "args": ["-y", "exa-mcp-server"],
    "env": { "EXA_API_KEY": "your-exa-key" }
  },
  "firecrawl": {
    "type": "stdio",
    "command": "npx",
    "args": ["-y", "firecrawl-mcp"],
    "env": { "FIRECRAWL_API_KEY": "fc-your-key" }
  },
  "fetch": {
    "type": "stdio",
    "command": "uvx",
    "args": ["mcp-server-fetch"]
  },
  "playwright": {
    "type": "stdio",
    "command": "npx",
    "args": ["-y", "@playwright/mcp"]
  }
}

Add only the servers you need. Each one starts a separate process, and its tools are added to every chat.

Hosted apps (HTTP)

Many services host their own MCP server at a URL. Use "type": "http":

Server	URL	Sign-in
notion	https://mcp.notion.com/mcp	OAuth in the browser
slack	https://mcp.slack.com/mcp	OAuth in the browser
canva	https://mcp.canva.com/mcp	OAuth in the browser
figma	https://mcp.figma.com/mcp	OAuth in the browser
hubspot	https://mcp.hubspot.com/anthropic	OAuth in the browser
github	https://api.githubcopilot.com/mcp/	GitHub token in a header

"notion": {
  "type": "http",
  "url": "https://mcp.notion.com/mcp"
},
"github": {
  "type": "http",
  "url": "https://api.githubcopilot.com/mcp/",
  "headers": { "Authorization": "Bearer <YOUR_GITHUB_TOKEN>" }
}

For OAuth servers, restart hrida-agent-sdk, run /mcp, select the server, and finish signing in through the browser.

Allow the tools

List every server you added in settings.json:

"permissions": {
  "allow": ["mcp__perplexity", "mcp__brave-search", "mcp__fetch", "mcp__notion"]
},

Servers that can change things

Allowing a server lets the agent use all of its tools without asking. For servers that can write, send, or delete (Slack, GitHub, Notion, HubSpot, Playwright), leave them out of allow so the agent asks before each action.

Add servers from the command line

Instead of editing ~/.hrida.json by hand, you can use hrida-agent-sdk mcp add-json with --scope user:

hrida-agent-sdk mcp add-json fetch '{"type":"stdio","command":"uvx","args":["mcp-server-fetch"]}' --scope user
hrida-agent-sdk mcp list

Troubleshooting

• perplexity is missing from /mcp: ~/.hrida.json isn't valid JSON, or the entry isn't inside mcpServers. Check for missing commas or brackets.
• perplexity shows as failed: npx isn't available or can't download the package. Run npx -y @perplexity-ai/mcp-server in the terminal to see the error.
• 401 Unauthorized / invalid API key: the key in PERPLEXITY_API_KEY is wrong or expired. Update it and restart.
• The agent asks for permission before every search: the permissions entry is missing or in the wrong file. Check step 2 and restart.