Skip to content

feat: document keyless Parallel Search MCP for custom agents #4946

Open
Open
feat: document keyless Parallel Search MCP for custom agents#4946
Labels
enhancementNew feature or requestmcpMCP integration changes

Description

@georgeatparallel
georgeatparallel
opened on Oct 11, 2026

Quick check

  • I've searched existing issues and didn't find a duplicate.
  • This relates to the SDK, CLI, or a specific agent.

What problem are you trying to solve?

GAIA developers need a runnable example for adding web search and page fetching to a custom agent without creating a search-service account or configuring an API key.

Proposed solution

Add an opt-in guide for Parallel Search MCP, which offers free anonymous search and page fetching, with search powered by Fast mode. Use GAIA's existing mcp_servers.json and MCPClientMixin, with a stdio-to-HTTP bridge such as mcp-remote connecting to https://search.parallel.ai/mcp. GAIA's client currently accepts stdio servers, so a URL-only configuration would not work.

Keep existing servers and web-tool defaults unchanged. Document the bridge dependency, setup and anonymous rate limits. This would be a custom-agent example rather than a catalog entry, since the catalog requires a built-in consumer. It would complement the web-search work tracked in #1460.

Acceptance criteria

  • A clean installation can load the documented configuration through GAIA's MCP manager and return useful search results and fetched page text without saved or environment credentials.
  • The custom-agent example dispatches the MCP tools through the agent loop; existing configuration and defaults remain intact.
  • The guide is linked in the documentation navigation, with focused configuration tests and live caller evidence.

Would this scoped documentation example fit GAIA?

I work at Parallel.

Activity

  1. github-actions
    added
    enhancementNew feature or request
    mcpMCP integration changes
    on Oct 11, 2026

  2. github-actions commented on Oct 11, 2026

    @github-actions
    github-actionsbot
    on Oct 11, 2026 – with GitHub Actions
    Contributor

    Thanks for the clear write-up, and for saying up front that you work at Parallel. A guide like this would fit an existing pattern, but adding a vendor-specific integration page to the docs is a maintainer's call, so I'm tagging @kovtcharov-amd to decide on scope.

    Your technical premises check out:

    • GAIA can only reach MCP servers it launches locally. A URL-only entry would be rejected, so you'd need a bridge like mcp-remote, and the guide should list its Node.js requirement.
    • There's already a slot for this. The MCP client docs include per-server walkthroughs (for example, Windows System Health), so a Parallel page could sit next to them as a custom-agent example. It wouldn't need to be a catalog entry.
    • Please keep it separate from EPIC: L2 Web Search & Extraction #1460. That epic covers GAIA's built-in web tools, and changing their defaults is out of scope here, as you noted.

    If the scope is approved, a PR would need a docs/docs.json nav entry, config tests that mock the bridge, and one live run showing a search result and a fetched page.

    🔍 Technical details
    • Stdio-only check: src/gaia/mcp/client/mcp_client_manager.py (~line 60) rejects any non-stdio type.
    • Existing per-server guide to model on: docs/guides/mcp/windows-system-health.mdx; client guide: docs/guides/mcp/client.mdx; spec: docs/spec/mcp-client.mdx.
    • Docs links inside src/gaia/ must keep the /docs/ prefix (enforced by tests/unit/test_amd_gaia_urls.py).
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or requestmcpMCP integration changes

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions