You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Feature]: Add a keyless Parallel Search MCP setup example #3453
I searched existing issues and docs before opening this.
This proposal is grounded in a real Roomote use case, not a broad generated feature list.
Problem
When a Roomote agent needs current library documentation or web context for a coding task, users should have a discoverable way to try web search and page fetching without provisioning a search API key. The custom MCP guide explains the transport and permissions, but has no concrete Parallel setup example.
Current workaround
Users can manually add https://search.parallel.ai/mcp as a remote, no-auth custom server through Integrations or Personal Settings → Personal MCP servers. This uses Roomote's existing custom MCP support.
Proposed behavior
Add a small, opt-in Parallel Search MCP example to the custom MCP guide: the endpoint, no-auth selection, an importable JSON snippet, and a sample search followed by page fetching. Search uses Fast mode with no account or API key required; the anonymous tier is intended for light use and has lower rate limits.
Keep existing integrations and tool permissions unchanged. If invited, I would validate importing the snippet, discovering tools, and executing search and fetch through Roomote's existing proxy before opening a PR.
Affected surfaces
Docs, MCP, Web dashboard.
Self-hosting implications
The proposal uses the existing remote custom MCP path. It would require outbound HTTPS access to search.parallel.ai, with no new service, port, environment variable, credential, storage, or migration.
Alternatives considered
Roomote already offers opt-in Exa search and fetching, including keyless access. This example would add a documented provider choice using existing custom MCP support. A built-in Parallel catalog preset could improve discoverability further, but a configuration example is the smaller starting point.
Documentation impact
Add the example to apps/docs/integrations/custom-mcp-servers.mdx.
I've analyzed this issue and recommend a small, docs-only implementation. The existing importer (apps/web/src/lib/custom-mcp-json-import.ts) accepts a URL-only mcpServers entry, and CustomMcpServers.tsx already provides Remote (HTTP) and Authentication → None. Parallel's current documentation describes anonymous access at /mcp and the web_search and web_fetch tools. No additional reporter clarification is needed to draft the example.
Proposed implementation plan
Add a “Keyless web search example: Parallel” section to apps/docs/integrations/custom-mcp-servers.mdx, near the existing import/setup guidance. Explain both entry points: shared under Integrations, or private under Personal Settings → Personal MCP servers. Show Import from JSON, confirm the remote URL and Authentication → None, choose the intended visibility, and save. Use this single-server snippet:
Include a short search → fetch walkthrough. For example, ask the agent to use Parallel to find current official documentation for a library, then fetch a returned documentation URL and summarize it with a citation. Name web_search and web_fetch; if showing tool inputs, use the provider's current schemas (objective and search_queries for search; urls for fetch). Explain how to confirm discovery in Manage tools, and use a newly started task when checking sandbox availability.
State the limits and prerequisites accurately. Anonymous use needs no account, API key, or auth headers and is intended for exploration/light use at lower rate limits. Parallel's default fast search mode is a provider setting, distinct from Roomote's Fast mode. Link to provider docs rather than inventing a fixed quota. Self-hosted deployments need outbound HTTPS access from the Roomote control plane to search.parallel.ai. Link to the existing Exa guide as another keyless option.
Validate before publishing. Import the exact snippet, confirm URL/no-auth prefill, discover both tools, and execute one search followed by fetching a returned public URL through Roomote's existing proxy. Verify private/shared visibility and existing per-tool controls behave as documented. Run mise exec -- pnpm --filter @roomote/docs check and inspect the rendered guide. Existing importer tests cover the configuration shape; additional application tests are only warranted if this check exposes a compatibility defect. Coordinate the edit with open docs PR [Docs] Explain custom MCP tool-list recovery #3451, which touches the same guide for tool-list recovery.
Acceptance criteria: a reader can copy the snippet, discover both tools, and complete the search/fetch example without credentials. The proposed change uses existing custom MCP support and preserves the catalog and tool-permission behavior; it requires no new service, environment variable, or migration.
This documents an opt-in provider choice without requiring a built-in integration. Live search/fetch through Roomote has not been exercised in this triage; it remains an explicit validation step. Maintainers can assess the docs-only proposal and reply with feedback; tag @roomote for follow-up.
Activity
roomote-community commentedon Oct 10, 2026
I've analyzed this issue and recommend a small, docs-only implementation. The existing importer (
apps/web/src/lib/custom-mcp-json-import.ts) accepts a URL-onlymcpServersentry, andCustomMcpServers.tsxalready provides Remote (HTTP) and Authentication → None. Parallel's current documentation describes anonymous access at/mcpand theweb_searchandweb_fetchtools. No additional reporter clarification is needed to draft the example.Proposed implementation plan
apps/docs/integrations/custom-mcp-servers.mdx, near the existing import/setup guidance. Explain both entry points: shared under Integrations, or private under Personal Settings → Personal MCP servers. Show Import from JSON, confirm the remote URL and Authentication → None, choose the intended visibility, and save. Use this single-server snippet:{ "mcpServers": { "parallel-search": { "url": "https://search.parallel.ai/mcp" } } }web_searchandweb_fetch; if showing tool inputs, use the provider's current schemas (objectiveandsearch_queriesfor search;urlsfor fetch). Explain how to confirm discovery in Manage tools, and use a newly started task when checking sandbox availability.fastsearch mode is a provider setting, distinct from Roomote's Fast mode. Link to provider docs rather than inventing a fixed quota. Self-hosted deployments need outbound HTTPS access from the Roomote control plane tosearch.parallel.ai. Link to the existing Exa guide as another keyless option.mise exec -- pnpm --filter @roomote/docs checkand inspect the rendered guide. Existing importer tests cover the configuration shape; additional application tests are only warranted if this check exposes a compatibility defect. Coordinate the edit with open docs PR [Docs] Explain custom MCP tool-list recovery #3451, which touches the same guide for tool-list recovery.Acceptance criteria: a reader can copy the snippet, discover both tools, and complete the search/fetch example without credentials. The proposed change uses existing custom MCP support and preserves the catalog and tool-permission behavior; it requires no new service, environment variable, or migration.
This documents an opt-in provider choice without requiring a built-in integration. Live search/fetch through Roomote has not been exercised in this triage; it remains an explicit validation step. Maintainers can assess the docs-only proposal and reply with feedback; tag @roomote for follow-up.