Parallel Search MCP and Dome
Agents describe what they want in plain language, and you can read it before it leaves. Parallel's Search MCP takes a natural-language objective and keyword queries, and fetches pages by URL. An objective often says more about your plans than any keyword. Behind Dome, a rule reads each objective before it leaves and a guard reads each keyword query.
Dome provides a governed gatewayfor Parallel.
Objectives read before they leave
An objective is a sentence about what the agent wants. A rule refuses one that names an internal project, searching or fetching.
Keyword lists guarded
Keyword queries arrive as a list. The guard reads every item, and a customer's email in one blocks the whole call.
Two tools, one key
Parallel's server has web_search and web_fetch. Dome holds the key for both, and each agent signs in to the Gateway with its own.
Parallel behind the Gatewayin four steps.
Add the Search MCP server, sync both tools, guard the keyword lists, then deploy the objective rule.
Add the Parallel Search MCP server
Dome stores the Parallel API key and sends it as a Bearer token on every call.
shelldome tools add --name parallel \ --url https://search.parallel.ai/mcp \ --auth-method api-key \ --credential-type shared \ --authorization "Bearer $PARALLEL_API_KEY" \ --gateway prod-gatewaySync the catalog
Sync needs a valid key, and Parallel refuses an invalid one. Without a sync, the Gateway has no Parallel tools to offer.
shelldome tools catalog sync parallelBlock personal data in arguments
A JSON filter on the request direction scans every argument. A match blocks the call before it reaches Parallel.
shellcat > query-leak.json <<'EOF'{"json":{"components":[{"fieldActions":[{ "matcher":{"contentMatchers":[{"ssn":{}},{"creditCard":{"requireLuhn":true}},{"email":{}}]}, "action":"FILTER_ACTION_BLOCK"}]}]}}EOFdome guards filters create query-leak --config-from query-leak.jsondome tools guards filters set parallel --direction request --filters query-leakApply the rules
The rule is scoped to deal-scout. Simulate an objective that names the codename, then deploy.
shelldome rules apply parallel-objectives.cedar --agent deal-scout --name parallel-objectives
Search and fetch,never the codename.
The permit opens both Parallel tools. The forbid refuses any call whose objective names Project Halcyon. Keyword queries arrive as a list, so the guard covers those.
cedar
permit (principal, action == Dome::Action::"mcp:call", resource is Dome::MCPTool)when { resource.connection_name == "parallel" && ["web_search", "web_fetch"].contains(resource.tool_name)};
forbid (principal, action == Dome::Action::"mcp:call", resource is Dome::MCPTool)when { resource.connection_name == "parallel" && resource has arguments && resource.arguments has objective && resource.arguments.objective like "*Halcyon*"};One call,two outcomes.
Switch the caller or the argument and watch the same call decide differently, with every decision landing in audit.
Objective
agent deal-scout · acting as m.chen
parallel/web_search(objective: "Find industrial sensor startups that raised a Series B in 2026", search_queries: ["industrial sensor startup Series B"])
- Agentdeal-scout is registered and active
- Callerm.chen verified through Okta
- RuleThe objective names no internal project
DecisionAllowed
Three agentson Parallel.
Each one tells Parallel what it's looking for. None of them tells it what the company is planning.
sourcing-agent
Find new suppliers
Searches for suppliers that meet a spec and reads their product pages for procurement.
support-researcher
Research a support case
Searches vendor docs and forums for a known error. A customer's email in a query blocks the call.
deal-scout
Scout acquisition targets
Describes the kind of company a deal needs in plain language. An objective that names the deal itself is refused.
Bringing it alltogether.
Connecting Parallel to registered agents, models, and identity in Dome completes a governed agent application.
Acting for
Control point
GatewayEvery call decided and audited
RulesGuardsQuotas
Tools
ParallelAgent service
Models
Common questionsabout Parallel.
More of whatDome works with.
Model
Claude Fable
Fable 5.1 from Anthropic and Amazon Bedrock in one failover pool, open to one group and capped by quota.
Provider
Anthropic
The Claude API behind the Model Broker: the key held in Dome, every call authorized, metered and audited.
MCP server
GitHub
The GitHub MCP server behind the Tool Gateway, with rules that decide each call on its owner and repository.
Identity
Microsoft Entra ID
Entra access tokens verified on every agent call, so rules read app roles and audit names the person.
Runtime
LangGraph
LangGraph agents with their model calls on the Model Broker and their MCP tools on the Tool Gateway.
Client
Claude Code
Claude Code on a Dome Gateway with per-developer sign-in, rules on every tool call, and audit by name.
Talk with ourFDE team.
Or start free, and register your first agent in minutes. Connect it through a gateway, give it rules, and see every call it makes in one audit trail.
Talk with our FDE team
Our engineers design, build and run governed agents for the part of your business you want to change. You see what they deliver.
Agent services: web, research, payments, voice and sandboxes
Web actions, research, payments, phone calls and code execution are where agents reach past your walls. Put each service behind the Gateway, and every call is authorized, metered and audited.