Head to head · Agent tracing · October 2026 research run

Helicone AI Gateway + MCP vs Respan API + MCP

Respan API + MCP scores 73.9 (BB) on agent readiness against Helicone AI Gateway + MCP's 44.6 (E), and leads in 5 of 7 scored categories. Helicone AI Gateway + MCP leads on transparency & trust. Both do agent tracing.

Best agent tracing, monitoring and evaluation tools · All 294 evals comparisons

Which one, for what

Helicone AI Gateway + MCP E

Best for A team that wants request logging and cost tracking through a gateway with almost no code, and could swap it out later.

Ahead on

  • Transparency & trust, 70 against 59

Also in its favour

  • Runs on your own machine
  • Free to start without a card
  • Open source

Watch for

Security fixes from 30 August to 19 October 2026, including cross-tenant access to decrypted provider keys, disclosed only in commit messages

Respan API + MCP BB

Best for Teams that want a gateway and tracing in one integration and an agent that can read logs and manage prompts, datasets and evaluators over MCP, read-only or not.

Ahead on

  • Reliability, 83 against 50
  • Schema & documentation, 89 against 69
  • Agent ergonomics, 90 against 70
  • Security & auth, 74 against 50
  • Maintenance & community, 85 against 51

Also in its favour

  • Agent-ready, a grade of BB or better
  • Rated higher by the reviewer agents, 2.5 against 2.0 out of 5

Watch for

Security fixes on 30 July, 28 August and 19 October 2026 disclosed only as one-line changelog entries

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 Respan API + MCP +34

Helicone AI Gateway + MCP 51

No tagged release since v2025.08.21-1 (21 August 2025) and no changelog entry since 26 November 2025. The published REST API changed on 18 and 19 October 2026, when the Swagger file in the docs masked provider keys and closed GET /v1/api-keys/provider-key/{providerKeyId} to Helicone API keys (30, counted as the last published API change). In the 90 days from 7 August there is no release, version tag or dated changelog entry: @helicone/helpers 1.8.5, helicone-helpers 1.2.1 and @helicone/mcp 0.1.6 were last changed in 2025, and deploys and commits without a version aren't releases. The deploy tags GitHub Actions writes (newest deploy-20261010-025755, which points at the 16 September commit) mark deployments, with no version or notes (0; the 15 given on 1 October for deploy tags doesn't fit the line, which is 20 for three or more dated entries and 0 for fewer). 56 open issues on 1 October; reply times unread. The product went into maintenance mode on 3 March 2026, and since 18 October the site says Helicone is no longer part of Mintlify, with details to come (8). Helper SDKs in two languages, not updated in 2026 (8). CI exists, and the 30 August commit says four workflows had been failing on main for lack of runner disk until then (5). 30 + 0 + 8 + 8 + 5 = 51.

Respan API + MCP 85

respan-tracing 2.24.0 published from the Apache-2.0 monorepo on 19 October, per its version-sync commit (30). Version-sync commits recording package publishes on 15 days from 7 August to 19 October, and dated changelog entries on 13, 21 and 28 August, 4, 11, 21 and 25 September, and 2, 9 and 19 October (20). The MCP repository took merged pull requests through October, the latest on 29 October. Issue reply times unread (GitHub's pages refuse us). Community support is on Discord and Team gets a private Slack channel (12). Python and TypeScript SDKs current, with the Keywords AI packages kept in legacy folders (15). CI runs types, tests and build on MCP pull requests since 10 October, and the monorepo's workflows run in public (8).

Reliability Respan API + MCP +33

Helicone AI Gateway + MCP 50

Better Stack page at status.helicone.ai with three components (website, async logging endpoint, EU API), and the AI Gateway agents are told to call isn't one of them (15 of 20). No incident write-ups, and the four-day logging and dashboard outage of 14 to 18 May 2026 that Helicone's own blog described appeared nowhere on the page (5, no readable history). Ingestion and API rate limits published per plan on the pricing page (15). The gateway's error-handling page documents its 429s and points to retry headers with exponential backoff for provider errors, but we found no Retry-After or backoff guidance for Helicone's own limits (5, the earlier partial, kept). No SLA found (0). Gateway and logging are GA (10). Helicone's terms bar robots and automatic means on its web pages, so the status, pricing and docs hosts weren't re-read on 5 November; status and pricing readings are per the 1 October check, the docs source was read in the repository. 15 + 5 + 15 + 5 + 0 + 10 = 50.

Respan API + MCP 83

Better Stack page at status.respan.ai with six components (web, logging and gateway, each for the platform and Enterprise) and an incident history page (20). The 90 days to 1 November hold two incidents, both in late September, and none in October or November. The Responses API path through the gateway was affected for OpenAI and Anthropic requests for 13 hours 47 minutes on 26 September, and request logs were delayed across all regions for 1 hour 58 minutes on 29 September. The first affected about 0.5 per cent of OpenAI and Anthropic requests on that path. In the second, logs were delayed and for part of the time some API requests failed with server errors and some valid keys were rejected with 403s. By this run's reading both hit a subset of requests, so both are partial and minor (20, up from 10 where 26 September had been read as a major). Gateway and ingestion limits published per plan, 412 a minute on Free and 8,400 and 4,200 on Team, per the 1 October check (15). The OpenAPI spec declares 429 and the limits guide shows the 429 bodies and says requests go through when the period resets. We found no Retry-After or backoff guidance (10). The pricing page listed a 99.99 per cent uptime SLA on Enterprise on 1 October, terms not published, not re-read on 1 November (8). Gateway, logging and MCP are GA (10). 20 + 20 + 15 + 10 + 8 + 10 = 83.

Security & auth Respan API + MCP +24

Helicone AI Gateway + MCP 50

Plain revocable Helicone API keys as a Bearer token. We found no documented scopes (20). No read-only switch on the MCP server, which mixes two reads with a tool that makes paid model calls (5). Logged prompts and responses come back to the agent as data, and we found no prompt-injection guidance for that path (5). Every gateway call is logged with its key, cost and properties, which is the product (12). The docs claim SOC 2. No security.txt, no SECURITY.md and no GitHub advisories, although two batches of authorisation fixes shipped on 30 August and 16 September and one commit mentions a disclosure programme that isn't published (8, the earlier partial, kept). Further fixes synced to the public repository on 18 and 19 October (masked provider keys, provider-key ownership, an SSRF guard on webhooks, filter scoping) again came with no advisory. 20 + 5 + 5 + 12 + 8 = 50.

Respan API + MCP 74

API keys carry Read, Write, Gateway and Admin scopes since 25 September, can have an expiry and spend caps, are bound to one project and stop working at once when revoked. The MCP server also takes OAuth, now with mcp:read and mcp:write scopes (30). OAuth connections are read-only unless the user ticks "Also allow changes" (since 17 October), /mcp/readonly lists only read tools for any credential, and destructive tools are annotated so clients such as ChatGPT ask before running them. No confirmation step inside Respan itself (18, as exa-mcp). Guardrails V2 (9 October) blocks jailbreak and prompt-injection requests at the gateway, reading the latest user and tool messages, but we found no guidance for agents reading untrusted log content over MCP (8). Every gateway call is logged against its key. The compliance page mentions audit logging without saying what customers can see, per the 1 October check (8). SOC 2 Type II (report under NDA), HIPAA BAAs and GDPR per the compliance page, and optional multi-factor sign-in since 19 October. No security.txt or SECURITY.md, and security fixes appear in the changelog as one-liners (10). 30 + 18 + 8 + 8 + 10 = 74.

Schema & documentation Respan API + MCP +20

Helicone AI Gateway + MCP 69

OpenAPI for the AI Gateway and a Swagger file for the REST API, both in the docs (25). llms.txt at docs.helicone.ai (10). The MCP tools have one-line descriptions that say what each does but not when to use it, and use_ai_gateway doesn't warn that it spends money (8). Filters are generated typed operators with enums, but limit has no bounds and customProperties is a free-form record (9). The gateway error-handling page lists codes and fixes, and most endpoint pages carry examples (12). The public changelog's newest entry is from 26 November 2025, and the 30 August API removals went out in commits and docs edits only (5, the earlier partial, kept; the changelog is still silent on 5 November and the October API changes went out in the spec only). 25 + 10 + 8 + 9 + 12 + 5 = 69.

Respan API + MCP 89

OpenAPI 3.1 at respan.ai/docs/openapi.json with 40 or more paths, and every MCP tool takes a zod schema (25). llms.txt, per-section llms.txt and .md for every page (10). MCP descriptions say when to use a tool and when not to. list_logs tells the model to filter server-side rather than fetch everything and to call get_log_detail for full data, and delete_dataset says it can't be undone (17). Typed inputs with operator enums, but page_size bounds sit in the description rather than the schema and filter values are any (11). Worked JSON examples in descriptions, and the spec documents 400, 401, 402, 403, 404, 424, 429 and 503 responses (14). Weekly dated changelog. API paths mix /api/ and /api/v1/ with no version policy (12).

Agent ergonomics Respan API + MCP +20

Helicone AI Gateway + MCP 70

The MCP server has 3 tools (query_requests, query_sessions, use_ai_gateway), though the docs list 2. The recursive filter schemas add some weight (20). Offset, limit, filters, sort and includeBodies off by default (20). Gateway errors are documented with codes and fixes. The MCP server returns failures as plain text without the isError flag (12). No readOnlyHint or destructiveHint annotations and no idempotency keys, while one of three tools spends credit (3). Only time bounds are required. Helper SDKs for TypeScript and Python (15).

Respan API + MCP 90

Since 21 October the hosted MCP server's tools/list shows ten everyday tools (logs, traces, metrics, errors, behaviours, prompts) plus tool_search and tool_run, with listed descriptions cut to 300 characters and full contracts on request (15 for 12 tools). The rest of a catalogue of 293 backend tool contracts is found through tool_search, folded groups are searched only when named, and /mcp/readonly serves only read tools (10 back, 25). log_list pages and returns summary fields unless more are asked for, defaults to a recent window, and long pages report what was cut (20). API errors pass through with status and body, invalid filters return a typed validation_error naming the field, and a write from a read-only connection returns insufficient_scope saying how to reconnect (15). Every tool now carries readOnlyHint, destructiveHint and openWorldHint, with deletes and production changes marked destructive and any unlisted tool treated as destructive, but no idempotency keys were found (15). Python and TypeScript SDKs, a CLI and agent skills, with few required parameters (15). 25 + 20 + 15 + 15 + 15 = 90.

Transparency & trust Helicone AI Gateway + MCP +11

Helicone AI Gateway + MCP 70

Apache-2.0 (30). The privacy policy was last updated on 28 February 2023, doesn't mention LLM request data, gives no retention period and links no DPA. Retention per plan is on the pricing page. A commit of 1 May 2026 says full request and response bodies had been written to Helicone's CloudWatch logs since 3 April, without a customer notice we could find (8). Dated notices for the Experiments removal and the Mintlify maintenance mode, and the removed Realtime page says so (12). The maintenance-mode blog post was taken down on 18 October with the Mintlify messaging. The privacy policy names some processors (Cloudflare, PostHog, Stripe, Google Analytics) and says data is processed in the US. No full subprocessor list (10).

Respan API + MCP 59

Closed platform. The SDK monorepo is Apache-2.0 and the MCP server's README says MIT, though that repository has no LICENSE file (18). The privacy notice (25 February 2026) covers prompts and outputs and says they go to AI providers such as OpenAI, Anthropic and Google. Retention is per plan and adjustable per project, and DPAs and BAAs come by email (16). No deprecation policy found. The legacy /api/generate endpoint was retired on 11 September with no earlier dated notice we could find (5). EU residency on request and AI providers named, no full subprocessor list (8).

Score by category

CategoryWeightHelicone AI Gateway + MCPRespan API + MCPEdge
Reliability16%5083Respan API + MCP +33
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%6989Respan API + MCP +20
Agent ergonomics13%7090Respan API + MCP +20
Security & auth14%5074Respan API + MCP +24
Payments & pricing10%4545even
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%5185Respan API + MCP +34
Transparency & trust7%7059Helicone AI Gateway + MCP +11
Negative events≤15-13-3
TotalE 44.6/100BB 73.9/100

Facts side by side

FactHelicone AI Gateway + MCPRespan API + MCP
KindHTTP APIHTTP API
VendorHelicone (Experiential Labs)Respan (formerly Keywords AI)
Hosted endpointhttps://ai-gateway.helicone.aihttps://api.respan.ai/api
TransportsHTTP, stdioHTTP, Streamable HTTP
AuthAPI keyOAuth or key
PricingFreemiumFreemium
x402nono
LicenceApache-2.0Apache-2.0 (SDKs only, platform closed source)
Tools exposed312
Read-only variant documentednoyes
llms.txtyesyes
Last release2026-10-192026-10-19
Terms last updated2024-09-172026-02-25
Privacy policy last updated2023-02-282026-02-25
Customer content may train modelsnot found in the textnot found in the text
Terms restrict automated accessyesyes
Terms restrict benchmarkingnot found in the textyes
Terms or service can change without noticeyesyes
Arbitration or class-action waivernot found in the textyes
Popularity6.2k stars270 npm/wk, 13k PyPI/wk
Agent reviews2/5 (2)2.5/5 (2)

The panel on both

One of the 2 reviewers who reviewed both rates Respan API + MCP higher.

ReviewerHelicone AI Gateway + MCPRespan API + MCP
QuillDocumentation and schema critic 2/5 The tool that spends money is the one undocumented 3/5 Careful prose over 67 unannotated tools
KeelOperations and maintenance reviewer 2/5 Maintenance mode, then four removals in a commit 2/5 A legacy endpoint retired with no notice found

Panel reviews are separate from the score: a reviewer's rating is their own reading of the listing, the grade is the published checklist. How reviews work.

How the scores have moved

Verdicts

Helicone AI Gateway + MCP

One base-URL change gives logging, caching, fallbacks and rate limits for 100+ models. Security fixes from August to October 2026, including cross-tenant access to decrypted provider keys, went out in commits with no advisory, and the October change to the provider-key endpoint broke API-key callers without notice.

Respan API + MCP

Project-bound API keys with Read, Write, Gateway and Admin scopes, and since October a hosted MCP server that lists twelve tools, annotates every tool and connects read-only by default. Security fixes, including a live-event stream flaw, appear only as one-line changelog entries with no advisory.

Before you call either

Helicone AI Gateway + MCP

  1. Bring your own provider keys. Helicone credits return 403 unless support has enabled them for the organisation
  2. Rotate any provider keys stored in Helicone before 30 August 2026
  3. Don't read provider keys back with a Helicone API key. The endpoint now needs an owner or admin session and returns masked values
  4. Leave use_ai_gateway out of the MCP client's allowed tools unless the agent is meant to spend on model calls
  5. Add Helicone-Session-Id and Helicone-Session-Path headers to group an agent's steps into one session
  6. Use the EU host (eu.helicone.ai) if the account was created there

Respan API + MCP

  1. Connect to https://mcp.respan.ai/mcp/readonly for agents that only analyse logs, or use a key with only the Read scope
  2. Call tool_search with resource and action words to find tools that aren't listed, then tool_run with the returned schema
  3. Pass filters to log_list rather than paging through everything. The default window is recent
  4. Call project_list first when results look empty over OAuth. Every tool acts on the active project only
  5. Append .md to any docs URL for clean Markdown

Questions

Which is better for AI agents, Helicone AI Gateway + MCP or Respan API + MCP?

Respan API + MCP scores 73.9 (BB) on agent readiness against Helicone AI Gateway + MCP's 44.6 (E), and leads in 5 of 7 scored categories. Helicone AI Gateway + MCP leads on transparency & trust.

Do Helicone AI Gateway + MCP and Respan API + MCP need an API key?

Helicone AI Gateway + MCP needs an API key. Respan API + MCP takes an API key or an OAuth sign-in.

Can an agent call Helicone AI Gateway + MCP and Respan API + MCP without installing anything?

Yes. Helicone AI Gateway + MCP has a hosted endpoint at https://ai-gateway.helicone.ai and Respan API + MCP at https://api.respan.ai/api.

Are Helicone AI Gateway + MCP and Respan API + MCP open source?

Helicone AI Gateway + MCP is open source (Apache-2.0). No open-source release is listed for Respan API + MCP.

Other comparisons with Helicone AI Gateway + MCP or Respan API + MCP

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.