ApiTap
Turn captured website traffic into reusable API skill files that agents can replay directly.
- Source repo
- n1byn1kt/apitap
- Stars
- ★ 132
- Last updated
- 1mo ago
- License
- Apache-2.0
- Primary language
- TypeScript
- FA score
- 75/100 · Good
At a glance
- How it runs
- Works with
- Universal · cross-platformClaude Code
- Cost
- Free, no paid service needed
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A developer integrating data from a site's single-page app can capture its traffic once and replay endpoints with fetch().
- Not a fit if
- Teams limited to public APIs and unable to rely on a site's internal endpoints
- Users whose red-tier anti-bot endpoints require full browser automation
- Website
- apitap.io ↗ set on GitHub
- Maintenance
- Commits in the last 6 monthslast commit 2026-10-21latest release v2.3.4 (2026-10-21)
- Source review
- 75/100 · Good
What does this agent do, and when should you use it?
ApiTap is a CLI, MCP server, and npm library for accessing website data through APIs instead of browser automation. It can capture live network traffic through Playwright/CDP, discover common APIs, or import OpenAPI/Swagger specs, then store the result as a signed JSON skill file on the local machine. During replay, ApiTap uses fetch() to call endpoints matched against their captured route shapes; credentials are injected from encrypted storage and kept out of skill files. It also offers browser-free peek, read, and discover commands, plus an optional Chrome extension. It requires Node.js 20+; capture and browser-based browsing also require Playwright Chromium, while the other paths do not need a browser.
Use apitap capture <url> or apitap attach --port <port> to observe site traffic and generate endpoint skill files; apitap discover <url> probes for APIs; and apitap import converts OpenAPI/Swagger specs from APIs.guru, a URL, or a local file. Skill files live in ~/.apitap/skills/ and describe routes, parameters, and response shapes; credentials are stored separately in encrypted storage. Run apitap replay <domain> <id> [key=val...] to replay an endpoint with fetch(), or use apitap read <url> to extract page content and apitap peek <url> for a HEAD check. MCP clients can call 12 tools, including apitap_read, apitap_capture, and apitap_replay.
- A developer integrating data from a site's single-page app can capture its traffic once and replay endpoints with fetch().
- An integration developer exploring public APIs can import an OpenAPI spec from APIs.guru and inspect its endpoint map.
- Users of Claude Code, Claude Desktop, Cursor, or Windsurf can configure ApiTap as an MCP server.
- An agent builder seeking to reduce HTML context can use
apitap readand its built-in site decoders to extract page content. - Users already signed into a site can use the optional Chrome extension to passively discover APIs or capture traffic from their browser session.
Key setup facts
Pulled from the install and usage notes below. The project README has the current values.
- Install / run
npm install -g @apitap/corenpx playwright install chromium
- Runtime
- Node.js 20+
How do you install or deploy this agent?
Requires Node.js 20+. Install the CLI and npm package globally:
npm install -g @apitap/coreAdd the MCP server to Claude Code:
claude mcp add -s user apitap -- apitap mcpInstall Chromium only if you want capture or browser-based browse:
npx playwright install chromiumread, peek, discover, and import work without a browser.
How do you use this agent?
Import Stripe's OpenAPI spec and replay an endpoint:
apitap import --from apis-guru --search stripe
apitap replay api.stripe.com get-listcharges limit=5To capture site traffic and then replay an endpoint, run:
apitap capture https://polymarket.com
apitap replay gamma-api.polymarket.com get-eventsBrowse the opened site normally during capture, then press Ctrl+C to finish. Skill files are saved under ~/.apitap/skills/. Replay depends on the endpoint being replayable; credentials, when needed, are stored encrypted after capture.
What are this agent's strengths and limitations?
- Offers three ways to build skill files: capture live traffic, discover APIs, or import OpenAPI specs.
- Replays endpoints with fetch() without launching a browser in the replay path; its MCP server exposes 12 tools.
- Keeps credentials out of skill files in separate encrypted storage, signs skill files, and matches replays to captured route shapes.
- Provides browser-free
peek,readdecoders, anddiscoveroperations.
- Requires Node.js 20+ and terminal access; MCP use also requires an MCP-compatible client.
captureand browser-basedbrowserequire a separate Playwright Chromium install, thoughattachcan connect to existing Chrome.- Replayability varies: CSRF or session-bound endpoints may be fragile, while signed or anti-bot endpoints may require a full browser.
- Skill files depend on capture or spec quality; the README gives imported endpoints confidence scores of 0.6–0.85, which may rise after a successful replay.
How does this agent compare with similar options?
Compared with Playwright/Puppeteer browser automation, ApiTap captures APIs first and then replays requests directly, avoiding DOM selectors and waits; complex anti-bot endpoints may still need a full browser.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| ApiTap This agent | 75 · Good | CLIFree | ★ 132 | 1mo ago | TypeScript | Claude Code |
| Scrapling Web Scraping Framework | 65 · Some gaps | Library / SDKFree | ★ 105k | 4d ago | Python | — |
| wigolo | 71 · Some gaps | CLIFree + model costs | ★ 5.6k | 11d ago | TypeScript | Codex · Claude Code · OpenAI API · Claude API |
| CamoFox Browser Server | 63 · Some gaps | Self-hosted serviceFree | ★ 452 | 15d ago | JavaScript | Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The materials describe SSRF checks, connection-time DNS validation, redirect rechecks, separation of credentials from skill files, PII scrubbing, signing, and restrictive file permissions. Capture can collect traffic from a real signed-in session and agents can replay requests; some credential refresh paths may open a browser. The extension's promotion flow mentions approval notification, but ordinary capture/replay has no consistent per-operation confirmation. Dependency security has SHA-pinned CI actions and npm installation guidance, but no dependency audit or vulnerability mitigation evidence. Cleanup and repair have quarantine, snapshots, and a restore path. Contributor and repository names are given, but publisher identity is unverified and maintenance ownership remains unclear.
Code and documentation largely explain discovery, capture, replay, and error handling; the CLI has a stated JSON error format. However, README, Hermes skill, and SECURITY materials are not fully aligned on versions, supported versions, and browser behavior. CI configuration shows npm ci and a test workflow, but static review cannot confirm execution results. Playwright is a hard dependency, and some environments need a separate Chromium install. Failure paths include structured errors and cause fields, but still depend on network targets and site behavior.
The product covers CLI, MCP, npm library, and Hermes use, with distinct read, peek, discover, import, capture, and replay paths. It explains browser requirements, replay tiers, and credential refresh conditions. Tool-selection guidance is fairly specific, though some entry points make broad applicability claims. Node 20+, Linux/macOS, and browser setup are documented; the heavy dependency and possible failure of auth refresh in sandboxes limit environment fit.
The README and two skill documents separately explain MCP and CLI usage, with installation, configuration, examples, FAQ-like comparisons, limitations, and a security model. Apache-2.0 metadata matches the license file. Package and Hermes skill versions exist, but the sample has no changelog and version mapping is unclear. A security reporting channel and response targets are stated, but responsibility is attributed only to ApiTap Contributors, without a clear maintainer. CLAUDE.md reads as Claude Code usage/contributor guidance rather than a product manifest.
JSON output, MCP tool tables, response budgets, and truncation markers support agent consumption. The main value is reusing website APIs, extracting structured content, and reducing browser costs. Token-savings and endpoint-confidence figures are supplied, but are mostly project claims in this sample; the Playwright hard dependency also raises installation cost. Benefits appear strongest on sites that are accessible and replayable.
README, code excerpts, package metadata, security policy, and sample tests jointly support some claims, including encryption, SSRF, and signing design. Most README figures and coverage claims are not independently substantiated by the supplied sample. The evidence distinguishes confidence levels, replay tiers, and unverified endpoints, and discloses encryption threat-model and scrubbing limits; performance and coverage claims lack comparably strong support.
- Credential protection and scrubbing are documented, but the security policy says encryption does not protect against an attacker who obtains the ~/.apitap directory and salt, and PII scrubbing is best-effort.
- Capture, the extension, and replay handle real session credentials or send requests to sites; review targets, permissions, and auth settings that may trigger browser refresh before use.
- Performance, token-savings, and test-count figures come from project materials; this static review did not execute the code or independently verify those figures.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
Hacker News listing pages read through the API, and Angular/Ember app shells are labelled as SPAs (PR #105). Found by the nightly dogfood run after its scorer was fixed to read apitap's own failure labels. Fixed - HN listings. /news, /best, /newest, /ask, /show and /jobs now read the matching Firebase list (top 10) instead of falling through to HTML extraction. Only / and item?id= were decoded before. Paging (?p=2) is ignored. The URL match is anchored, so a URL that only ends in an HN path is not decoded as an HN listing. …View on GitHub ↗ -
Release notes excerpt
read and peek now say when they could not get the page instead of returning an empty success, and OpenAPI import keeps a spec's port (PRs #102, #103; issues #96–#101). Found by the nightly dogfood run over 180 public URLs: 18 previous failures now pass, 0 regressions. Security - Import keeps credentials on their port (#96). Stored auth is keyed by hostname. Re-importing a spec served on another port of the same host used to move the skill file's base URL there, so stored credentials would follow. An existing skill file now keeps its origin …View on GitHub ↗ -
v2.3.2 v2.3.2 — per-endpoint query keys; no auth on https→http redirects
Release notes excerpt
Security follow-ups to v2.3.1's query-key support, found reviewing PR #89 (PRs #94, #95; issues #81, #90, #91, #92, #93). Security - Per-endpoint query keys (#92). Stored query keys were one value per parameter name per domain, so a domain that used a read-only key on one endpoint and an admin key on another replayed every endpoint with whichever was captured first. Each key is now tagged with the endpoint it was captured on (method + parameterized path), and replay sends an endpoint the key captured on it. If only one key is stored it is used …View on GitHub ↗
FAQ
Does ApiTap require a paid API or model key?
Which features work without a browser?
read, peek, discover, and import do not need a browser. capture and browser-based browse need Playwright Chromium; attach can connect to an already running Chrome.Do skill files contain login credentials?
Can every captured endpoint be replayed directly?
Is ApiTap open source?
Is ApiTap safe to use?
Is ApiTap still maintained?
Who makes ApiTap?
What is ApiTap's official website?
Thanks — we use this to improve the page.