Head to head · Exchange market data · October 2026 research run

Bitget vs Bitpont API and MCP

Bitpont API and MCP scores 62.9 (B) on agent readiness against Bitget's 56.4 (C), and leads in 4 of 7 scored categories. Bitget leads on agent ergonomics, maintenance & community and transparency & trust. Both do exchange market data.

Best crypto exchange APIs and MCP servers for AI agents · All 231 exchanges comparisons

Which one, for what

Bitget C

Best for An agent trading spot, futures and earn products in a person's Bitget account outside the prohibited list, with the smallest default MCP surface of the exchanges graded and a paper-trading switch.

Ahead on

  • Agent ergonomics, 83 against 74
  • Maintenance & community, 78 against 14
  • Transparency & trust, 65 against 58

Also in its favour

  • Runs on your own machine

Watch for

Hot and warm wallets lost about 388 million dollars on 24 September 2026 to forged withdrawal commands, and withdrawals were paused until 2 October

Bitpont API and MCP B

Best for An aggregator over DeFi venues (Jupiter, KyberSwap, LI.FI, deBridge, Aave, Morpho, Hyperliquid, GMX, Derive) rather than an exchange with its own order book.

Ahead on

  • Reliability, 45 against 40
  • Schema & documentation, 82 against 74
  • Payments & pricing, 90 against 60

Also in its favour

  • A hosted endpoint, with nothing to install
  • Free to start without a card
  • No incidents deducted, where Bitget loses 8 points for them

Watch for

No status page, no SLA and no sandbox or testnet in the reviewed documentation

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 Bitget +64

Bitget 78

Read as a hosted service with an open-source toolkit. The UTA changelog's latest launched entry is 30 October 2026 and a preview entry 3 November; the MCP package shipped 3.3.1 on 20 September and agent-skill was pushed 30 October (30). 58 dated changelog entries since 5 August 2026 and six MCP versions since June (20). Issue trackers on the Bitget-AI repositories (11 open on agent_hub) and a support centre; replies were not checked (10 of 25). In the official MCP registry under io.github.Bitget-AI, a GitHub namespace, with official SDKs in five languages (10 of 15). CI workflow with coverage, npm provenance attestations, two runtime dependencies, no install scripts (8 of 10). 78.

Bitpont API and MCP 14

Read as a closed hosted service. The changelog names versions (Bitpont 5.102.20, API 1.9) with no dates, so the time since the last release can't be established (0) and there are no dated entries (0). A public changelog and a support email; no answer observed and the GitHub organisation is closed to our reader (6 of 15). A first-party JavaScript client and TypeScript declarations as downloads, not on a package registry, and no MCP registry entry (8 of 15). No package or public CI seen (0). The public announcement was on 20 November 2026, so the record is days old. Total 14.

Payments & pricing Bitpont API and MCP +30

Bitget 60

No machine payment protocol (0). Base spot and futures rates are stated in a help article without login, the tier table draws by script (20). Market data needs no key, demo trading costs nothing and no card is asked for at sign-up (20). A person registers, verifies identity and mints the key in a browser (0). 40. Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 40 is 60.

Bitpont API and MCP 90

No x402, MPP or L402; a verified trading fee lifting quotas for an hour is not a payment protocol (0). The API is free under published limits and the 0.40% interface fee is published per route without a login (30). Free access with no card (30). An agent starts with no signup by creating a private workspace through the API (30). Total 90.

Agent ergonomics Bitget +9

Bitget 83

Read for the API with the MCP server as the agent route. 14 compact tools by default, module filters, a full surface on request and progressive discovery (25). Cursor pagination on CFD order history since 15 October 2026 and limit parameters on lists (15 of 20). Uniform ok or error result shape from the SDK with typed error classes and compact JSON in the text channel (13 of 20). clientOid on orders, dryRun: true on every write, confirm: true gating cancelAll and withdraw, annotations derived from risk level including readOnlyHint, destructiveHint and idempotentHint (18 of 20). Few required fields, official SDKs in five languages plus the TypeScript agent SDK and a CLI (12 of 15). 83.

Bitpont API and MCP 74

Read for the hosted MCP server, whose catalogue has 1,148 tools (5), with 10 added back for ?product= profiles, bitpont_findTools and bitpont_describeTool and paginated tools/list (15 of 25). Cursors, limits, product and network filters and read batches (18 of 20). Errors carry code, retryable and requestId, and checkInputs returns JSON Pointer paths before any provider is called (19 of 20). Idempotent tasks and stable request IDs, with a documented rule never to repeat an unknown submission; readOnlyHint and destructiveHint unchecked because tools/list was not sent, and the catalogue carries no annotations (14 of 20). A JavaScript client with TypeScript declarations, OpenAI function definitions and a Postman collection, one language (8 of 15). Total 74.

Schema & documentation Bitpont API and MCP +8

Bitget 74

The agent SDK ships openapi.yaml with 109 typed UTA v3 operations from which the tools are generated, a machine-readable contract from the vendor's own organisation though not on the docs site (20 of 25). llms.txt in the SDK and MCP repositories; the docs site has none (6 of 10). Intent tools with a discover tool that explains each verb, action and its risk level, and the server's initialize instructions state surface, modules and read-only posture (15 of 20). Typed inputs with enums and required fields from the OpenAPI source (12 of 15). Examples in the quick start and READMEs, a REST error code page in the legacy docs that we did not read in full (9 of 15). Changelog by day with preview notices and effective dates (12 of 15). 74.

Bitpont API and MCP 82

OpenAPI 3.1 file Bitpont Developer API 1.33.0 with 1,149 operations, read once as the reference viewer's source (25). llms.txt, Markdown downloads of the developer pages and Markdown agent guides (10). Every operation has a description and most a summary that states its job, and the guides say which tool to use for which task, though few say when not to (14 of 20). Enums, regex patterns and length limits on addresses and base-unit amounts, required fields; some inputs still allow free-form objects and the MCP operation takes a generic params object (12 of 15). Code examples in the guides and a shared error schema with code, requestId and retryable on 400 to 504 (13 of 15). A versioned changelog, but its entries carry no dates and its latest API version (1.9) doesn't match the file's 1.33.0 (8 of 15). Total 82.

Transparency & trust Bitget +7

Bitget 65

The editorial half. A closed hosted service with clear terms naming BTG Technology Holdings Limited; the toolkit is MIT (15 of 30). The privacy notice names the controller and EEA representative, gives 7-year and up to 10-year retention for compliance and litigation, and names recipients by category only (18 of 30). Deprecations and parameter changes carry effective dates (delta mode deprecated 27 August, symbol and category required from 30 September, snapshot changes with demo and live dates) but no written policy (15 of 20). Recipients by category, transfers under Seychelles law with no locations or provider names (5 of 20). 53.

Bitpont API and MCP 58

The editorial half. Closed source under readable terms, which licence the published SDK and documentation for integrations (15 of 30). A detailed privacy notice naming the operator, bases, recipients and browser storage, but it promises no deletion period for server logs and support records and no DPA was found (18 of 30). No deprecation policy or dated notices; the changelog only says changes are additive (5 of 20). Integration providers named by function (Jupiter, deBridge, Velocity, Jito, Kamino, KyberSwap, LI.FI, 1inch); the hosting provider and data locations are given only on request (10 of 20). Total 48.

Reliability Bitpont API and MCP +5

Bitget 40

Graded with the hosted lines on the UTA v3 API at api.bitget.com. No status page found (status.bitget.com does not resolve, the docs link none), so 0 for the page and 5 for no readable history; the September 2026 breach is in the deduction, not here. Rate limits are published as 6,000 per IP per minute and per-endpoint UID limits by VIP level (15). An x-mbx-used-remain-limit header is documented and the agent SDK backs off on 429, orders take a clientOid of up to 64 characters, but the breach error code and a Retry-After rule were not read (10 of 15). No SLA (0). UTA v3 is the production API (10). 40.

Bitpont API and MCP 45

Read with the hosted lines on the REST API and hosted MCP server. No status page linked from any page read (0). No readable incident history (5). Rate limits with numbers, 30 request units a minute per IP for free access, 60 for partner keys, six items a batch (15). RateLimit and Retry-After headers documented, a retryable flag on every error, and stable request IDs and saved tasks for recovering a write after a lost response (15). No SLA (0). Nothing in the reviewed pages marks the API as beta or preview (10). 0 + 5 + 15 + 15 + 0 + 10 = 45.

Score by category

CategoryWeightBitgetBitpont API and MCPEdge
Reliability16%4045Bitpont API and MCP +5
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%7482Bitpont API and MCP +8
Agent ergonomics13%8374Bitget +9
Security & auth14%6263Bitpont API and MCP +1
Payments & pricing10%6090Bitpont API and MCP +30
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%7814Bitget +64
Transparency & trust7%6558Bitget +7
Negative events≤15-80
TotalC 56.4/100B 62.9/100

Facts side by side

FactBitgetBitpont API and MCP
KindHTTP APIHTTP API
VendorBTG Technology Holdings LimitedBitpont (LiMiT)
Hosted endpointno (local only)https://www.bitpont.com/api/v1/mcp
TransportsstdioStreamable HTTP
AuthAPI keyOAuth or key
PricingPay per usePay per use
x402nono
LicencenoneA proprietary hosted service under the Bitpont Terms of Use (LIMIT SRL). The terms licence the published SDK, declarations, examples and API documentation for building integrations, with no copyright fee; the backend is closed
Tools exposed141148
Read-only variant documentedyesno
llms.txtnoyes
MCP registryio.github.Bitget-AI/bitget-agent-mcpnot listed
Last release2026-10-30none
Terms last updated2026-10-20no date given
Privacy policy last updated2025-12-052026-11-11
Customer content may train modelsnot found in the textnot found in the text
Terms restrict automated accessyesnot found in the text
Terms restrict benchmarkingnot found in the textnot found in the text
Terms or service can change without noticeyesnot found in the text
Arbitration or class-action waiveryesnot found in the text
Popularity234 starsnone

How the scores have moved

Verdicts

Bitget

Bitget's official MCP server keeps the default surface to 14 intent tools with discovery, read-only and paper-trading switches, dry runs and a confirmation gate on withdrawals, over an API with demo trading. A 388 million dollar wallet breach in September 2026 paused withdrawals for eight days, and there is no status page, SLA or readable fee table.

Bitpont API and MCP

Public market reads, quotes and unsigned transaction drafts work without an account, every trade is signed in the owner's wallet, and an OpenAPI 3.1 file covers all 1,148 operations. There is no status page, the changelog carries no dates, and the hosted MCP server exposes the same 1,148 tools, narrowed by product filters and a search tool.

Before you call either

Bitget

  1. Run npx -y @bitget-ai/bitget-agent-mcp --read-only for research and --paper-trading with a demo key before any live order; the two flags cannot be combined
  2. Call discover({}) first, then the intent verbs; add dryRun: true to any write to preview it, and pass confirm: true only after the person has agreed to a withdraw or cancelAll
  3. Send a unique clientOid of at most 64 characters on every order and reuse it on retry; orderId wins when both are sent
  4. Stay under 10 requests a second per endpoint on a new account and 6,000 a minute per IP; the SDK retries on 429, so do not add a second retry loop
  5. Bind the key to an IP and keep write permission off a research key; the docs read show no separate withdrawal scope

Bitpont API and MCP

  1. Connect with ?product=swap or another product ID, then call bitpont_findTools and bitpont_describeTool rather than loading all 1,148 tools
  2. Create a private workspace before the first ticket and store its secret and your requestId; reuse both to recover an order
  3. Call bitpont_checkInputs before preparing a transaction; amounts are base-unit decimal strings
  4. If a request returns 429, wait for the Retry-After value. The free budget is 30 units a minute per IP and a batch costs one unit per item
  5. Never repeat a submission whose outcome is unknown; read the execution receipt first

Questions

Which is better for AI agents, Bitget or Bitpont API and MCP?

Bitpont API and MCP scores 62.9 (B) on agent readiness against Bitget's 56.4 (C), and leads in 4 of 7 scored categories. Bitget leads on agent ergonomics, maintenance & community and transparency & trust.

Do Bitget and Bitpont API and MCP need an API key?

Bitget needs an API key. Bitpont API and MCP takes an API key or an OAuth sign-in.

Can an agent call Bitget and Bitpont API and MCP without installing anything?

Bitget runs on your own machine, with no hosted endpoint listed. Bitpont API and MCP has a hosted endpoint at https://www.bitpont.com/api/v1/mcp.

Other comparisons with Bitget or Bitpont API and 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.