Nimble

By Nimble

Get New App Alerts

Open in ChatGPT

Open in Claude

Overview

Nimble gives agents governed, accurate access to the public web, at a lower token cost per answer. Search finds current pages with ranked results and snippets. Extract returns clean, structured content from any URL, including pages needing JavaScript or interaction. Extraction Templates give consistently shaped records from known sites. Web Search Agents handle open-ended work across many sources, returning findings with citations, for research, enrichment, and dataset building. Also included: ready-made workflows for competitor research, company profiles, meeting prep, market discovery, brand monitoring, SEO, and talent sourcing.

Tools

nimble_agent_templates_get

ChatGPT

Get a Web Search Agent gallery template. Not an Extract Template.

nimble_agent_templates_list

ChatGPT

List Web Search Agent gallery templates to seed nimble_agents_create(template=...). These are not Extract Templates — do not pass names to nimble_extract_templates_*. Prefer a matching gallery template over a free-form create.

nimble_agents_create

ChatGPT

Create a Web Search Agent. Prefer reuse: list existing, then gallery templates. Pass `template

 (from nimble_agent_templates_list) when a gallery blueprint matches; otherwise describe the agent via name/goals/sources. Classify

use_case`: research = dossiers/briefings/synthesis; dataset_building = entity lists/catalogs/extract-all without a URL; enrichment = add fields to records the user already has. Set effort from the task — do not ask the user to pick: low/medium simple, high default, x-high/max complex; dataset_building requires high or above. For a one-shot, prefer nimble_agents_run(agent_name=...) create-or-reuse instead of create-then-run.

nimble_agents_delete

ChatGPT

Delete a Web Search Agent permanently.

nimble_agents_get

ChatGPT

Get a Web Search Agent's skill, sources, effort, use_case, and output schema. Workspace agents only — do not pass an Extract Template name or a gallery template name here (use nimble_agent_templates_get for gallery templates).

nimble_agents_list

ChatGPT

List Web Search Agents in the account — reuse before creating a new one. Use for research, enrichment, dataset building, named-subject gathering, "extract all …" without a URL, catalog/directory lists, and fan-out beyond 1–few known URLs. Classify use_case: research = dossiers/briefings/synthesis; dataset_building = entity lists/catalogs/extract-all without a URL; enrichment = add fields to records the user already has. Then: get → run → run_status (poll) → run_result. If none fit, check nimble_agent_templates_list before free-form create.

nimble_agents_run

ChatGPT

Start a Web Search Agent run (research / enrichment / dataset building). Returns immediately with a run `id and status. Poll nimble_agents_run_status until terminal, then nimble_agents_run_result. Provide agent_id to run an existing agent, or agent_name to create-or-reuse by name (the one-shot path). Classify use_case` when creating: research = dossiers/briefings/synthesis; dataset_building = entity lists/catalogs/extract-all without a URL; enrichment = add fields to records the user already has. Set effort from the task — do not ask the user to pick: low/medium simple, high default, x-high/max complex; dataset_building requires high or above.

nimble_agents_run_result

ChatGPT

Fetch a completed Web Search Agent run (text or JSON plus trust/citations). Call after nimble_agents_run_status reports a terminal status.

nimble_agents_run_status

ChatGPT

Check a Web Search Agent run. Poll every ~15-30s until terminal, then run_result.

nimble_agents_runs_list

ChatGPT

List past and in-progress runs of a Web Search Agent.

nimble_agents_update

ChatGPT

Update an existing Web Search Agent via JSON Patch operations.

nimble_crawl_list

ChatGPT

List crawl jobs, optionally filtered by status.

nimble_crawl_run

ChatGPT

Start a web crawl to extract content from multiple pages on a website. The crawl discovers and visits pages starting from the given URL, following links up to the specified limit. Results can be retrieved via nimble_crawl_status. Latency & when to use: Asynchronous and cold-start-bound. This call returns a crawl_id immediately, but the job is queued: the first pages typically don't begin completing until ~1-2 minutes after dispatch, and total time scales with limit. Poll nimble_crawl_status for progress. Use this only when you need many pages from a site discovered and extracted automatically — it is the slowest of the extraction paths to first result. For a single known URL you want inline, use nimble_extract; for one URL where you can wait and poll, use nimble_extract_async.

nimble_crawl_status

ChatGPT

Check the status and progress of a running or completed crawl job.

nimble_crawl_terminate

ChatGPT

Cancel a running or queued crawl job.

nimble_extract

ChatGPT

Extract and parse content from a specific URL using Nimble's Extract API. This tool fetches and extracts clean, structured content from any web page. Use when you have a specific URL and need its full content. Latency & when to use: Synchronous — returns the content in this single call. Typical latency is sub-second to a few seconds for standard pages, so this is the default choice for a single URL you need inline right now. JavaScript-heavy or single-page-app pages take longer to render and can run up to the tool-call timeout configured by your MCP client (commonly ~180s) (especially when wait is set); a page that requires a login or returns a consent/interstitial prompt may yield little or no content. If you have many URLs to extract, or your turn's latency budget can't absorb a slow page, use nimble_extract_async instead. For discovering and pulling many pages across a whole site, use nimble_crawl_run.

nimble_extract_async

ChatGPT

Start an asynchronous URL extraction. Returns immediately with a task ID. Use this for batch processing or when you don't need results immediately. Retrieve results later with nimble_task_results using the returned task ID. Latency & when to use: This call returns a task ID in sub-second time, but the extraction itself runs out-of-band: a single URL typically takes ~30s end-to-end before results are ready (queueing and dispatch overhead make the async path an order of magnitude slower than the synchronous nimble_extract on comparable pages — it is not a faster route to the same content). After starting, poll nimble_task_results with the returned task ID, or supply callback_url to be notified on completion. Prefer this path when extracting many URLs in parallel, or when a single synchronous call would blow your turn's latency budget. For one URL you need inline now, use nimble_extract. For crawling many pages across a site, use nimble_crawl_run.

nimble_extract_templates_generate

ChatGPT

Generate a new Extract Template — only when the user explicitly asks to. Do not use as the no-template fallback. If list/get found no match, route to a Web Search Agent (nimble_agents_*) instead of generating. Returns a `generation_id; poll nimble_extract_templates_generation_status until terminal (completed or failed`). On completion, run via nimble_extract_templates_run.

nimble_extract_templates_generation_status

ChatGPT

Check the current status of a generation. Status flow: - in-progress (`pending / processing / etc.): call again after ~30 seconds. - completed: template_name and version_id are set; proceed to run. - failed / error: inspect error` for diagnostics.

nimble_extract_templates_get

ChatGPT

Get an Extract Template's input/output schema before running it. Use after nimble_extract_templates_list. This is an Extract Template, not a Web Search Agent gallery template (nimble_agent_templates_get).

nimble_extract_templates_list

ChatGPT

Browse Extract Templates — reusable, site-specific structured scrapers. First step when a named site plus a direct item (URL or site-specific ID) is known. Not the default for open research, enrichment, or dataset building. If none matches, use a Web Search Agent (nimble_agents_*) — do not generate a new template and do not dump a raw Extract. Rows are summaries (name, metadata, required_params). Full input_schema, output_schema and samples come from nimble_extract_templates_get. Typical workflow: list → get (inspect input_schema) → run (1–few known items).

nimble_extract_templates_run

ChatGPT

Run an Extract Template against 1–few known items (URL or site ID). Params must match the template's input_schema — inspect with nimble_extract_templates_get. Do not loop this to discover coverage; that fan-out is a Web Search Agent. If no results are returned, say so: the URL may be unreachable or behind a login wall, the page structure may have changed, or required params may be missing. Recover in this capability (fix params / URL) before switching class.

nimble_extract_templates_run_async

ChatGPT

Start an asynchronous Extract Template run. Returns a task ID immediately. Same capability as nimble_extract_templates_run, out of band: poll nimble_task_results. For 1–few known items only — not a substitute for a Web Search Agent fan-out.

nimble_extract_templates_update_from_template

ChatGPT

Refine an existing Extract Template — only when the user explicitly asks to. Not the no-template fallback (that is a Web Search Agent). Returns a `generation_id; poll nimble_extract_templates_generation_status until terminal, then run with the returned template_name`.

nimble_map

ChatGPT

Discover all URLs on a website by crawling its pages and sitemap. Returns a flat list of URLs found on the site. Useful for understanding site structure before targeted extraction or crawling.

nimble_search

ChatGPT

Nimble Search: raw web results for a single closed lookup. Use only when one search is enough — who/what at a glance, a filing or page pointer, a reverse lookup, or named-channel presence. Not for research, enrichment, dataset building, named-subject information gathering, or "extract all …" — those are a Web Search Agent (nimble_agents_*). Named site + a direct item → check Extract Templates first. For a closed synthesized fact, set include_answer=true (Answer); still not a substitute for a Web Search Agent.

nimble_task_results

ChatGPT

Get the status and results of an async task. Use this to retrieve results from nimble_extract_async or nimble_extract_templates_run_async. If the task is still pending, returns the current state so you can poll again later.