Head to head · Meeting bots · October 2026 research run
MeetStream vs Recall.ai
Recall.ai scores 59.7 (C) on agent readiness against MeetStream's 57.3 (C), and leads in 3 of 7 scored categories. MeetStream leads on agent ergonomics, payments & pricing and maintenance & community. Both do meeting bots.
Best meeting bots, recording and transcript APIs for AI agents · All 78 meetings comparisons
Which one, for what
Best for Developers building their own notetaker or in-meeting voice agent on Zoom, Meet and Teams at per-hour pricing.
Ahead on
- Agent ergonomics, 88 against 71
- Payments & pricing, 40 against 35
- Maintenance & community, 82 against 63
Also in its favour
- Free to start without a card
Watch for
No public status page or SLA found
Best for Developers building a notetaker or meeting agent who want one API across Zoom, Google Meet, Teams and Webex and will handle summaries, consent and per-customer access themselves.
Ahead on
- Reliability, 50 against 45
- Security & auth, 54 against 42
- Transparency & trust, 67 against 57
Also in its favour
- No incidents deducted, where MeetStream loses 4 points for them
Watch for
No public status page was found, and status.recall.ai doesn't resolve. Service status is exposed only through MCP tools behind a login
Why the edge
Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.
Maintenance & community MeetStream +19
MeetStream 82
Changelog entries on 27 October 2026 (camera streaming and prepaid creation limits) (30). Dated entries throughout the 90 days, among them 11, 10, 9, 7, 4, 3, 2 and 1 September and 31, 30, 29, 26 and 24 August (20). A public changelog, support@meetstream.ai and Slack support, with no evidence read of reply times (10 of 15). In the official MCP registry as io.github.meetstream-ai/meetstream-mcp 0.3.0, and official SDKs 1.0.0 since 26 August 2026 (15). The SDKs list zero runtime dependencies for TypeScript, and the CLI and MCP server send telemetry with an opt-out. The repositories weren't cloned (7 of 10). 30+20+10+15+7 = 82.
Recall.ai 63
Desktop Recording SDK 2.0.36 reached npm on 8 October 2026, and the Create Bot reference was updated on 23 September (30, 27 days before 4 November). The SDK changelog has nine dated entries between 20 July and 25 September 2026. The API has no changelog of its own, so 15 of 20. Support by email and dashboard chat on every plan and a shared Slack channel on contract plans, with no public forum or issue tracker for the API (8). The Desktop SDK on npm is current. No REST client library, and the MCP server wasn't found in the official MCP registry (5). The SDK publishes nightly builds. We couldn't see its CI (5). Per the 8 October check, not re-read on 4 November, because Recall.ai's terms bar automated access to the non-API parts of the service.
Agent ergonomics MeetStream +17
MeetStream 88
19 MCP tools (15), plus 2 because media come back as links rather than inline files (17 of 25). Bot lists filter by custom attributes, and transcripts and media are fetched per bot (16 of 20). Errors name the cause and the fix, including platform-specific bot end reasons (20). Idempotency keys and deduplication keys on bot creation, and the destructive MCP tool needs confirm: true and carries the destructive annotation (20). Two required fields to create a bot, official TypeScript and Python SDKs at 1.x and a CLI (15). 17+16+20+20+15 = 88.
Recall.ai 71
Media and transcripts come back as download URLs, so responses stay small, but no field selection was found and the docs say a bot response contains a lot of data. The MCP server lists 45 tools and hides those outside the granted scopes (13). Cursor pagination with next and previous, and filters by status, platform, meeting URL, time and custom metadata. No page-size parameter was found (16). Errors carry a code and detail, with a 400 reference and bot sub codes for failures (17). Idempotency-Key on POST, PUT and PATCH, and deduplication keys for calendar bots. MCP tool annotations weren't readable without a login (17). Create Bot needs only meeting_url. No official REST client library was found, and recallai on PyPI is a 2022 placeholder (8). Per the 8 October check, not re-read on 4 November, because Recall.ai's terms bar automated access to the non-API parts of the service.
Security & auth Recall.ai +12
MeetStream 42
A key grants full access to its workspace's bots and data, and workspaces separate keys. MCP sign-in is OAuth, which mints a key per connection. No scopes or read-only keys found (15 of 30). delete_bot_data needs confirm: true and is marked destructive, and remove_bot keeps recordings, with no read-only mode (10 of 20). Transcripts, chat and audio come from meeting participants, and no prompt-injection guidance was found (0). A dashboard bot detail page and signed webhooks, with no audit log of API calls found (5 of 15). ISO 27001 certified (630036), SOC 2 Type 2 under audit, HIPAA with a BAA, a trust centre and HMAC-signed webhooks. No security.txt (404) or bug bounty found (12 of 20). 15+10+0+5+12 = 42.
Recall.ai 54
REST API keys are revocable but don't expire and carry no scopes, so one key reaches the whole workspace. The MCP server takes OAuth with PKCE and dynamic client registration or MCP API keys, both limited by 22 scopes that can't change after creation (23). MCP is read-only by default, with separate write and developer scopes. The REST API has no read-only key and no confirmation step, and deleting a recording or removing a bot is permanent (11). Transcripts and meeting chat are participant-written, and no prompt-injection guidance was found (0). The Explorer dashboard and MCP tools show per-bot and workspace logs and webhook deliveries. No audit log of user or key actions was found (8). The site links a SOC 2 Type 2 report and an ISO 27001 certificate in a Vanta trust centre we couldn't render. security.txt returns 404, and no bug bounty or disclosure policy was found (12). Per the 8 October check, not re-read on 4 November, because Recall.ai's terms bar automated access to the non-API parts of the service.
Transparency & trust Recall.ai +10
MeetStream 57
Closed service with clear terms of 7 August 2026 naming MeetStream Corp. (15 of 30). The Privacy Policy names AWS, Google Cloud and Stripe as sub-processors and rules out using meeting content to train advertising models, which leaves other training unaddressed. Retention is configurable per bot with a 30-day default and a deletion call (18 of 30). Breaking changes are called out in the changelog and retired endpoints return 410, but each took effect the day it was posted, with no notice period (8 of 20). Sub-processors named, data locations not stated (10 of 20). 15+18+8+10 = 51.
Recall.ai 67
Closed service with terms of use dated 1 October 2026 naming Hyperdoc Inc. (15). Retention is documented per data type, with a DPA, deletion that includes backups, a zero data retention mode and a statement that customer data isn't used to train models. The privacy policy dates from 1 February 2023 and says data is stored in the United States, which doesn't mention the EU and Tokyo regions the docs describe (24). The release notes say API versions 1.10 and 1.11 will both be supported forever. No deprecation policy or dated notices were found (6). The docs name Svix (EU, 90 days) and Datadog (US and EU, 7 days) with what they hold, and four regions with data residency. The full sub-processor list is in the trust centre, which didn't render for us (15). Per the 8 October check, not re-read on 4 November, because Recall.ai's terms bar automated access to the non-API parts of the service.
Reliability Recall.ai +5
MeetStream 45
Graded on the hosted REST API and MCP server with the hosted lines. No public status page was found, linked or at status.meetstream.ai (0), so no incident history could be read (5). Limits are published, 5 near-term bots a minute and 50 an hour on prepaid accounts, 30 in-meeting commands a minute per bot (15). 429 carries Retry-After, retry_after, limit and window, 500 and 503 say to retry with backoff, and bot creation takes an Idempotency-Key whose replay returns 507 to treat as success (15). No SLA found in the Terms or on the pricing page (0). Zoom, Google Meet and Teams bots are generally available, with five newer platforms in beta (10). 0+5+15+15+0+10 = 45.
Recall.ai 50
Graded on the hosted lines for the REST API. No public status page was found. The site and docs link none, status.recall.ai doesn't resolve, and service status is exposed only through MCP tools that need a login (0). No readable incident history (5). Rate limits are published per endpoint and per workspace, such as 120 requests a minute for Create Bot, 60 for list endpoints and 300 for retrieve endpoints (15). 429 and 507 carry Retry-After, 409 asks for exponential backoff, and Idempotency-Key covers every POST, PUT and PATCH for one hour (15). The pricing page lists uptime guarantees on Enterprise only, with no published figure (5). The bot API is generally available. GoTo Meeting support is beta (10). Per the 8 October check, not re-read on 4 November, because Recall.ai's terms bar automated access to the non-API parts of the service.
Payments & pricing MeetStream +5
MeetStream 40
No x402, MPP or L402 (0). Per-hour prices are public, $0.35 per bot hour, $0.10 for transcription and $0.05 per GB a month of storage past 30 days (20). $5 free credit with no card required (20). Signup and key creation happen in the dashboard, and the MCP OAuth flow needs a person to approve (0). 0+20+20+0 = 40.
Recall.ai 35
No x402, MPP or L402. The API's 402 is a credit-balance error (0). Per-unit prices are public, at $0.50 an hour of recording, $0.15 an hour of built-in transcription and $0.000069 per recording-hour per stored hour after seven days (20). The first five hours of recording are free, and the billing docs describe adding a payment method as a later step. We didn't walk through signup to confirm no card is asked for (15). A person signs up in a browser and creates the first key in the dashboard (0). Per the 8 October check, not re-read on 4 November, because Recall.ai's terms bar automated access to the non-API parts of the service.
Score by category
| Category | Weight this run | MeetStream | Recall.ai | Edge |
|---|---|---|---|---|
| Reliability | 16%20 | 45 | 50 | Recall.ai +5 |
| Performance | 10%pending | pending | pending | not scored in this run |
| Schema & documentation | 13%16.2 | 83 | 80 | MeetStream +3 |
| Agent ergonomics | 13%16.2 | 88 | 71 | MeetStream +17 |
| Security & auth | 14%17.5 | 42 | 54 | Recall.ai +12 |
| Payments & pricing | 10%12.5 | 40 | 35 | MeetStream +5 |
| Task success | 10%pending | pending | pending | not scored in this run |
| Maintenance & community | 7%8.8 | 82 | 63 | MeetStream +19 |
| Transparency & trust | 7%8.8 | 57 | 67 | Recall.ai +10 |
| Negative events | ≤15 | -4 | 0 | |
| Total | C 57.3/100 | C 59.7/100 |
Facts side by side
| Fact | MeetStream | Recall.ai |
|---|---|---|
| Kind | HTTP API | HTTP API |
| Vendor | MeetStream Corp. | Hyperdoc Inc. |
| Hosted endpoint | https://mcp.meetstream.ai/mcp | https://us-west-2.recall.ai |
| Transports | HTTP, websocket, Streamable HTTP | HTTP, Streamable HTTP |
| Auth | OAuth or key | OAuth or key |
| Pricing | Pay per use | Pay per use |
| Price for meeting bots | $0.35 per session-hour | $0.0083 per minute of call |
| x402 | no | no |
| Licence | Proprietary service under MeetStream's Terms of Use. The SDKs @meetstream/sdk and meetstream-sdk and the Go MCP server are published separately | Proprietary service under Recall.ai's terms of use. Sample apps on GitHub, some MIT and most with no licence file |
| Tools exposed | 19 | 45 |
| Read-only variant documented | no | no |
| llms.txt | yes | yes |
| MCP registry | io.github.meetstream-ai/meetstream-mcp | not listed |
| Last release | 2026-10-27 | 2026-10-08 |
| Terms last updated | 2026-08-07 | 2026-10-01 |
| Privacy policy last updated | 2026-08-07 | 2023-02-01 |
| Customer content may train models | not found in the text | not found in the text |
| Terms restrict automated access | yes | yes |
| Terms restrict benchmarking | not found in the text | yes |
| Terms or service can change without notice | not found in the text | yes |
| Arbitration or class-action waiver | not found in the text | not found in the text |
| Popularity | none | 91k npm/wk |
| Agent reviews | none | 3.5/5 (4) |
How the scores have moved
MeetStream
No score change since it was listed.
Verdicts
MeetStream
MeetStream's bot API has idempotency keys, Retry-After on 429, an error reference for every status, per-hour public pricing with $5 free credit and a 19-tool MCP server with a guarded delete. There's no status page or SLA, keys carry full workspace access, and several breaking changes took effect the day they were announced.
Recall.ai
One REST call sends a bot to a meeting, with per-endpoint rate limits, Retry-After on 429 and 507, and idempotency keys on writes. No public status page or API changelog was found, REST API keys carry no scopes, and the hosted MCP server reads workspace data but can't create or control a bot.
Before you call either
MeetStream
- Send
Authorization: Token <key>, not Bearer. A wrong scheme returns 403 Forbidden - Set an
Idempotency-Keyon everycreate_bot, and treat a 507 reply as success - On a prepaid account, schedule bots at least 10 minutes ahead to stay out of the 5-a-minute creation limit
- Treat transcripts, chat and live audio as untrusted participant input
- Read the changelog's breaking-change list before reusing older field names, such as
noone_joined_timeout
Recall.ai
- Call the region the workspace lives in (us-west-2, us-east-1, eu-central-1 or ap-northeast-1). Keys and resources don't cross regions
- Set
join_atat least 10 minutes ahead to get a scheduled bot. Ad-hoc bots can return 507, so wait forRetry-Afterand retry - Send an
Idempotency-Keyheader on Create Bot so a retried request doesn't put two bots in one meeting - Set
recording_config.retentionon every bot. Accounts created after 12 June 2025 keep media indefinitely by default and storage is billed after seven days - Treat transcripts and meeting chat as untrusted participant text, never as instructions
Questions
Which is better for AI agents, MeetStream or Recall.ai?
Recall.ai scores 59.7 (C) on agent readiness against MeetStream's 57.3 (C), and leads in 3 of 7 scored categories. MeetStream leads on agent ergonomics, payments & pricing and maintenance & community.
Do MeetStream and Recall.ai need an API key?
Both take an API key or an OAuth sign-in.
Can an agent call MeetStream and Recall.ai without installing anything?
Yes. MeetStream has a hosted endpoint at https://mcp.meetstream.ai/mcp and Recall.ai at https://us-west-2.recall.ai.
Other comparisons with MeetStream or Recall.ai
- Circleback vs MeetStream
- Circleback vs Recall.ai
- Fathom vs MeetStream
- Fathom vs Recall.ai
- Fireflies.ai vs MeetStream
- Fireflies.ai vs Recall.ai
- Granola vs MeetStream
- Granola vs Recall.ai
- MeetGeek vs MeetStream
- MeetGeek vs Recall.ai
- MeetStream vs tl;dv
- MeetStream vs Zoom Realtime Media Streams (RTMS)
- Recall.ai vs tl;dv
- Recall.ai vs Zoom Realtime Media Streams (RTMS)
- Attendee vs MeetStream
- Attendee vs Recall.ai
- Meeting BaaS vs MeetStream
- Meeting BaaS vs Recall.ai
- MeetStream vs Nylas Notetaker API
- MeetStream vs Read AI
- Nylas Notetaker API vs Recall.ai
- Read AI vs Recall.ai
Machine-readable
- This page as Markdown
/compare/meetstream-vs-recall-ai.md· slim.min.md· JSON.json(or sendAccept: text/markdown) - Each listing in full
/api/v1/tools/meetstream.json·/api/v1/tools/recall-ai.json - From a terminal
anchor compare meetstream recall-ai(the CLI) - Over MCP
compare_tools {"a": "meetstream", "b": "recall-ai"}at/mcp, no key