Bland AI API + MCP vs Inworld Realtime API

Bland AI API + MCP 65

No x402 or L402. The headless signup endpoint answers 402 with a Machine Payments Protocol challenge settled through Stripe, but only to buy the $29.99 Agent Phone Plan, not to pay for calls (10 of 40). Per-minute rates for every self-serve plan are public, re-read on 2 November (20). The Start plan needs no card and comes with 2 credits and an inbound number (20). An agent can open an account through the headless flow, with the owner's only step being to relay a 6-digit texted code, or through a device-code flow that sends the owner to a browser (15 of 20). 10 + 20 + 20 + 15 = 65.

Inworld Realtime API 40

No x402, MPP or L402 (0). No per-session price, but each billed part (STT per hour, TTS per character, LLM per token) is priced publicly (20). On-Demand includes free STT and TTS allowances and lists no payment method, though some premium LLMs need a card on file (20). A person signs up in a browser before a key exists (0). Total 40.

Bland AI API + MCP 79

Organisation API keys, several per organisation and each revocable on its own, and since the 1 October check restricted keys with per-resource scopes (calls:read, calls:create, agents:write and others) and OAuth client credentials with the same permissions for short-lived tokens (an organisation that doesn't see OAuth clients in its dashboard is told to contact support). Restricted keys are rolling out by organisation and aren't available on pinned deployments, credited as conditional availability elsewhere in this run (30). A key can now be limited to read scopes, sign-in connections to the MCP server take mcp:read or mcp:write scopes, and destructive tools carry an annotation for the client's approval prompt. The server no longer asks for confirmation itself, and personas, numbers and tools stay full-access only (15 of 20). Agents read caller speech and transcripts, which are untrusted, and we found guard rails mentioned in the CLI but no prompt-injection guidance (5 of 15). Audit logs exist but are an enterprise function and don't record API key use, and every call has a log; the key list shows who created each key (12 of 15). security.txt valid until 2027-04-05 (re-read on 2 November), a trust portal, SOC 2 Type II and a PCI DSS assessment by a QSA per the vendor, and a disclosure address. No bug bounty found (17 of 20). 30 + 15 + 5 + 12 + 17 = 79.

Inworld Realtime API 58

Workspace API keys sent as Basic credentials, deleted to revoke, with separate keys per environment advised. The CLI mints read-only keys unless --write is given, but write toggles cover only the Voice and Router APIs, so no key is limited to one product. Browsers use short-lived tokens minted server side. Session IDs, not secrets, go in the URL (25 of 30). Tools run in the client, so the developer approves side effects, with no approval step in the API (10 of 20). The agent listens to callers and passes their speech to an LLM, and no prompt-injection guidance was found (5 of 15). Usage per key and a Performance page with errors and throttling (8 of 15). SOC 2 Type II and HIPAA claimed and a vulnerability report form, with no security.txt or bug bounty, per the 4 November reading (10 of 20). Total 58. Zero data retention covers session content from the Developer plan up, but not requests routed to third-party LLM providers.

Bland AI API + MCP 53

The hosted MCP server lists 62 dedicated tools and 282 API Reference tools, one per reference page, plus 3 prompts on key connections. Sign-in connections see a smaller allowlist, but there are no toolsets or dynamic loading (5 of 25). List endpoints such as audit logs and knowledge bases page and filter, but we found no field selection on call logs (14 of 20). Errors carry codes and readable messages, and a restricted key's 403 names the permission it lacks (16 of 20). Every MCP tool carries a read, write or destructive annotation, but there are no idempotency keys on call creation (10 of 20). A call needs only phone_number and a task, agent_id, persona_id or pathway_id, with defaults for voice and temperature. Official tooling is the Node CLI, a web SDK and a TTS Node library, and we found no Python SDK (8 of 15). 5 + 14 + 16 + 10 + 8 = 53.

Inworld Realtime API 68

Scored as an API. Conversation history defaults to the last 100 items, can be trimmed, and can be cached on Anthropic models, and every response.done carries a usage block (18 of 25). No lists to page. Output controls are modalities, turn detection settings, optional transcription and response.speak (12 of 20). Error events follow OpenAI's type and code shape, unsupported fields come back as non-fatal unsupported_field warnings, and item_not_found names the bad parameter (14 of 20). Tools run on the client and a follow-up response starts automatically after a tool result. Nothing covers side effects of a tool call cut off by an interruption (12 of 20). Works with OpenAI Realtime clients after changing the URL and key, with no Inworld Realtime SDK (12 of 15). Total 68.

Bland AI API + MCP 72

Closed source with clear terms (15). The privacy policy (last updated 23 June 2026) keeps data 'as long as necessary' with no periods for recordings or transcripts, and the trust page says zero-retention processing is available. The terms and the privacy policy both name Intelliga Corporation, doing business as Bland AI (12 of 30). The December 2025 price change was announced with a date and transition credits, and the migration from pathways and personas to agents carries a 15 November 2026 deadline in the docs, but we found no deprecation policy (8 of 20). A subprocessor list of 17 entries with locations, updated 8 September 2026, per the 1 October check, not re-read on 2 November (20). 15 + 12 + 8 + 20 = 55.

Inworld Realtime API 63

Closed service with clear terms naming Theai, Inc. dba Inworld AI (15). The ZDR page says what a session keeps and that third-party LLM requests aren't covered, but the terms and privacy notice disagree on training (12 of 30). Release notes flag potentially breaking changes with the clients affected and what to do, without stating a notice period (14 of 20). US headquarters with EU and India residency on Enterprise, and no sub-processor list in the docs, while sessions may route to third-party LLM providers (10 of 20). Total 51.

Bland AI API + MCP 68

No OpenAPI document found. The MCP server's dedicated tools take typed inputs and its API Reference tools take each reference page's parameters, but no machine-readable contract is published (5 of 25). llms.txt with 551 entries, each page served as Markdown (10). MCP tool descriptions say when to use a tool, for example the reference tells agents to use send_call and to choose one of agent_id, persona_id, pathway_id or task (14 of 20). The REST reference types every parameter with enums and defaults, though pathways travel as a raw graph and call_bland_api is a free-form passthrough (11 of 15). Errors for 400, 401, 402, 403, 404, 429 and 500 are documented with example messages (13 of 15). Versioned v1 and v2 APIs, agent versions and environments, and a dated changelog (15). 5 + 10 + 14 + 11 + 13 + 15 = 68.

Inworld Realtime API 60

No AsyncAPI or other machine-readable file for the socket, and the spec paths redirect (0). llms.txt, Markdown twins and a docs MCP server (10). The overview, OpenAI migration guide, providerData reference and the billing page's account of how a session is counted say what to use when (16 of 20). The reference lists every client and server event with field types and required marks, as text rather than a schema (12 of 15). JavaScript and Python example projects and an error event with type, code and param, without a list of codes (9 of 15). Dated release notes with affected clients and required actions, but no API version beyond /v1 (13 of 15). Total 60.