Integration

Parallel Web Systems integration

Parallel is a web research API built for AI agents. Search the web and images, extract page content from URLs, answer typed questions grounded in web evidence, run asynchronous research tasks (single or in task groups) that return structured, cited output, build verified lists of companies or people with FindAll runs, find entities in real time, watch the web with scheduled monitors, get cited answers through an OpenAI-compatible Responses route, and read or prune the account's research memory. Every create call is billed per request or per run (prices depend on mode, processor or generator; see the pricing page). Streaming event routes, the beta chat completions route and the legacy v1beta search and extract routes are not exposed. Authentication is an API key from the Parallel platform settings, sent in the x-api-key header.

Connect Parallel Web Systems How integrations work

Ask it in chat

Say what should happen.
The workflow builds itself.

When this happens, do that in Parallel Web Systems.Built from the 33 Parallel Web Systems tools below, on the connection you add once.

What you can do with Parallel Web Systems

Each of these is one operation on one endpoint, available both as a workflow step and as a tool an AI agent can call on its own. Nothing bundles several operations into one action, so what a run did stays readable afterwards.

Most endpoints run on the free plan. A few, mainly the ones that publish on your behalf, need a paid plan.

Monitor, 5 operations

cancel_monitorCancel monitorPOSTPermanently stops a monitor. Cancellation cannot be undone (create a new monitor to resume); cancelling an already cancelled monitor changes nothing. Returns the monitor.
create_monitorCreate monitorPOSTCreates a monitor that runs on a schedule and records an event when web content changes materially: event_stream watches a search query, snapshot re-checks a task run's output. Each execution is billed (lite $3, base $10 per 1,000).
get_monitorGet monitorGETReads one monitor: type, status (active or cancelled), frequency, processor, settings, last run time and, for snapshot monitors, the latest snapshot output.
list_monitorsList monitorsGETLists monitors, newest first, active only unless status is set (snapshot output is read with get_monitor). Pass next_cursor from the response as cursor for the next page; there are no more pages when it is absent.
update_monitorUpdate monitorPOSTChanges a monitor's frequency, processor, webhook, metadata or (event_stream only) settings. Only fields sent are changed; leave the others out to keep them. Returns the monitor.

Findall run, 4 operations

cancel_findall_runCancel findall runPOSTCancels a running FindAll run; it stops generating and evaluating candidates. Returns no body.
create_findall_runCreate findall runPOSTStarts a FindAll run that finds and verifies entities matching all match_conditions, and returns it with status queued. Poll get_findall_run, then read matches with get_findall_run_result. Billed by generator.
extend_findall_runExtend findall runPOSTRaises a FindAll run's match limit by additional_match_limit so it keeps finding matches; each extra match is billed at the generator's per-match price. Returns the run's updated spec.
get_findall_runGet findall runGETReads a FindAll run's status (queued, action_required, running, completed, failed, cancelling, cancelled), its candidate metrics and its termination_reason.

Findall spec, 2 operations

generate_findall_specGenerate findall specPOSTTurns a natural-language objective into a suggested FindAll spec (entity_type, match_conditions, generator, match_limit) to review and pass to create_findall_run. Does not start a run.
get_findall_run_schemaGet findall run schemaGETReturns the spec a FindAll run uses: objective, entity_type, match_conditions, enrichments, generator and match_limit.

Memory, 2 operations

clear_memoryClear memoryPOSTDeletes every entry from the selected research memory (personal memory, or the scope named by memory_scope_key). The underlying task runs, monitors and FindAll runs are kept. Cannot be undone. Returns no body.
retrieve_memoryRetrieve memoryPOSTRetrieves past task runs, monitors and FindAll runs from research memory, ranked by relevance to query, or the most recent ones when query is empty, with input and output excerpts.

Task group, 2 operations

create_task_groupCreate task groupPOSTCreates an empty task group to batch and track many task runs. Returns taskgroup_id; add runs with add_task_group_runs and follow progress with get_task_group.
get_task_groupGet task groupGETReads a task group's aggregate status: number of runs, counts per run status, whether any run is still active, and a status message.

Task group run, 2 operations

add_task_group_runsAdd task group runsPOSTStarts up to 1,000 task runs in a task group in one call and returns their run_ids. Each run is billed by its processor when it completes. Read each output with get_task_run_result.
get_task_group_runGet task group runGETReads the status of one run in a task group, same as get_task_run. Read its output with get_task_run_result.

Task run, 2 operations

create_task_runCreate task runPOSTStarts an asynchronous research task run and returns it at once with status queued and its run_id. Read the output with get_task_run_result (waits for completion) or poll get_task_run. Billed per completed run by processor.
get_task_runGet task runGETReads a task run's status (queued, action_required, running, completed, failed, cancelling, cancelled) without its output. Read the output with get_task_run_result.

Decision, 1 operation

make_decisionsMake decisionsPOSTAnswers 1 to 100 typed questions (choice, score or probability) about a shared context, grounded in current web evidence, and returns each answer with the web sources used. Billed $1 per 1,000 requests.

Entity, 1 operation

search_entitiesSearch entitiesPOSTReturns a ranked list of companies or people (name, url, description) matching a natural-language objective, in about 1 to 3 seconds. Billed $5 per 1,000 requests (100 results included).

Findall enrichment, 1 operation

add_findall_enrichmentAdd findall enrichmentPOSTAdds an enrichment to a FindAll run: extra fields researched for each matched candidate, billed per match at the chosen Task processor's price. Returns the run's updated spec.

Findall run result, 1 operation

get_findall_run_resultGet findall run resultGETReturns a snapshot of a FindAll run: the run status and every evaluated candidate with its match_status (generated, matched, unmatched, discarded), output and citations.

Image, 1 operation

search_imagesSearch imagesPOSTSearches the web for images and returns up to 20 results with direct image URLs and source pages. Billed per request: fast $1, advanced $5 per 1,000 requests.

Memory entry, 1 operation

evict_from_memoryEvict from memoryPOSTRemoves one task run, monitor or FindAll run from the selected research memory, so later runs no longer use it as context. The resource itself is kept. Returns no body.

Monitor event, 1 operation

list_monitor_eventsList monitor eventsGETLists a monitor's detected events, newest first, with each event's output and changed fields. Pass next_cursor as cursor for more, or event_group_id for one execution.

Monitor run, 1 operation

trigger_monitor_runTrigger monitor runPOSTRuns a monitor once now, outside its schedule; the execution is billed like a scheduled one and records an event only if it finds a material change. Cancelled monitors cannot be triggered. Returns no body.

Monitor stats, 1 operation

get_monitor_statsGet monitor statsGETCounts the account's monitors: the total, and per status and processor the number of monitors at each run frequency.

Response, 1 operation

create_responseCreate responsePOSTAnswers a question with live web research and URL citations, in OpenAI Responses format. reasoning.effort sets depth and price: low $10, medium (default) $50, high $250 per 1,000 answers. Waits for the answer.

Task run input, 1 operation

get_task_run_inputGet task run inputGETReturns the input a task run was created with (processor, input, task_spec, source_policy, settings and metadata). MCP server details and webhook URLs are not returned here.

Task run result, 1 operation

get_task_run_resultGet task run resultGETWaits until a task run completes and returns its output (text or JSON content with citations per field) and the run. Answers 408 if still running after timeout, 404 if the run failed or does not exist.

Web page, 1 operation

extract_urlsExtract urlsPOSTExtracts content from up to 20 web pages as markdown excerpts focused on an objective, or as full page content when enabled in advanced_settings. Billed $1 per 1,000 URLs.

Web search, 1 operation

search_webSearch webPOSTSearches the web and returns ranked page results with markdown excerpts sized for an LLM. Billed per request: turbo/fast $1 and basic/advanced $5 per 1,000 requests (10 results included).

Use Parallel Web Systems from Claude Code, Cursor or Codex

Connect Parallel Web Systems in LiveContext, create an API key in Settings, then add the LiveContext MCP server to your client. Its Parallel Web Systems operations become tools the client can call, next to every other integration you connected.

Claude Code

claude mcp add --transport http livecontext https://livecontext.ai/mcp --header "X-API-Key: YOUR_API_KEY"

Codex (~/.codex/config.toml)

[mcp_servers.livecontext]
url = "https://livecontext.ai/mcp"

[mcp_servers.livecontext.http_headers]
"X-API-Key" = "YOUR_API_KEY"

Cursor, Claude Desktop and other JSON configs

{
  "mcpServers": {
    "livecontext": {
      "type": "http",
      "url": "https://livecontext.ai/mcp",
      "headers": {
        "X-API-Key": "YOUR_API_KEY"
      }
    }
  }
}

Set up the MCP server step by step

How to connect Parallel Web Systems

Create your workspace
Sign up free. Nothing to install.
Connect Parallel Web Systems once
Add your Parallel Web Systems connection in Settings. Every tool from this integration reuses it, and secrets are encrypted at rest and visible only to you.
Describe the job in chat
The workflow builds itself in front of you and picks the endpoints it needs. Drag the steps yourself on the canvas whenever you would rather be explicit.