Head to head · PDF conversion · October 2026 research run

PDF.co API + MCP vs PDFShift

PDFShift scores 66.4 (B) on agent readiness against PDF.co API + MCP's 40.9 (E), and leads in 5 of 7 scored categories. PDF.co API + MCP leads on schema & documentation. Both do pdf conversion.

Best PDF manipulation and conversion APIs for AI agents · All 67 pdf comparisons

Best document parsing, OCR and extraction APIs for AI agents · All 218 documents comparisons

Which one, for what

PDF.co API + MCP E

Best for An agent that needs many PDF operations from one vendor and can handle files by URL.

Ahead on

  • Schema & documentation, 74 against 69

Also in its favour

  • Runs on your own machine

Watch for

No status page. status.pdf.co redirects to an unpublished Better Stack page

PDFShift B

Best for Agents that turn HTML or a live page into a PDF or screenshot in one call, with no browser to run.

Ahead on

  • Reliability, 73 against 20
  • Security & auth, 56 against 23
  • Payments & pricing, 75 against 30
  • Maintenance & community, 77 against 64

Also in its favour

  • Free to start without a card
  • No incidents deducted, where PDF.co API + MCP loses 3 points for them

Watch for

Keys come in two levels (master and basic) with no read-only scope, and the MCP server exposes delete_template with no tool annotations

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.

Reliability PDFShift +53

PDF.co API + MCP 20

status.pdf.co still answers its robots.txt request with a 302 to incidents.betterstack.com marked unpublished-status-page=true (not followed), so there's no status page (0) and no readable incident history (5). The June 2026 changelog moved rate-limit windows from per second to per minute, and on 10 November 2026 neither the request-limits page nor the getting-started page gives the numbers (0). The response-codes page defines 429 as too many requests and 503 as overloaded, with no Retry-After or backoff guidance. The MCP server's 0.0.10 tip for 429 says only to wait a moment and retry. 5 of 15 held for the documented code and the async job and webhook-retry route. No SLA in the terms of 24 March 2026, which reserve the right to change or stop the service without notice (0). GA (10).

PDFShift 73

Read with the hosted lines. status.pdfshift.io is a Better Stack page with eight monitors (API and two processors in each of Gravelines and Strasbourg, the website and the dashboard) (20). Its incident list shows no incidents from September to November 2026, but the Strasbourg API monitor shows 99.660 per cent uptime, over a window the page doesn't state, so downtime went without a write-up; read as minor incidents only (20). Rate limits with numbers: 250 conversions a minute authenticated, 10 a minute in sandbox, 2 a minute without a key, 50 parallel conversions (15). 429 is documented with X-RateLimit-Remaining, X-RateLimit-Limit and X-RateLimit-Reset, but no Retry-After, backoff advice or idempotency guidance (8). No SLA on any listed plan; the home page mentions one only for custom Enterprise plans (0). GA (10). 20 + 20 + 15 + 8 + 0 + 10 = 73.

Payments & pricing PDFShift +45

PDF.co API + MCP 30

No x402, MPP or L402 (0). Plan prices are public (Basic $8.99 a month billed yearly to Business 3 at $270, unchanged on 10 November 2026) and the credit cost of each endpoint is published per page or per call, so per-unit cost is readable without a login (20). The pricing FAQ asks what happens after the free trial and the terms say a cancelled account reverts to the free tier, but no page we read states the trial's size or whether a card is needed, so the line stays absent (0). A person signs up in a browser to get a key (0). Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 20 is 30.

PDFShift 75

No x402, MPP or L402 (0). Plan prices with a per-credit overage for every plan, effective per-document rates and a public pricing endpoint, all without a login (30). Free plan of 50 credits a month with no card, and free watermarked sandbox runs (30). Judgement call on onboarding. The API converts without a key, so an agent can start with no signup, but the output is watermarked, capped at 2 MB and 2 requests a minute; production use needs a browser signup and the key arrives by email (15 of 30). 0 + 30 + 30 + 15 = 75.

Security & auth PDFShift +33

PDF.co API + MCP 23

One API key per account in the x-api-key header, header only per the getting-started page and the OpenAPI document. We found no scopes, no multiple keys and no rotation guide (10). Every MCP tool also takes api_key as an argument, which invites the key into the model's context. No read-only mode, no annotations, and upload_file reads any absolute local path and uploads it to PDF.co. The SSE and HTTP transports still bind 0.0.0.0 with no authentication in 0.0.11 (0). PDF text, tables and invoice fields come back as untrusted content with no injection guidance (0). Each response reports credits used and remaining, which is a per-call cost trail. We found no audit log (5). The security page claims SOC 2 Type II and cites AWS's ISO 27001, PCI and FISMA, which are the data centre's; the DPA of 30 September 2026 promises breach notice within 48 hours. security.txt was a 404 on 1 October, not re-read, and we found no disclosure policy or bounty (8).

PDFShift 56

Named keys per account, master or basic, each disabled and re-enabled through the API, with no expiry or finer scopes; the key travels in a header and no query-string option is documented (24). Basic keys can't manage keys, but there is no read-only key, and the MCP server runs delete_template with no annotation or confirmation (6). Results are download URLs, logs and the account's own templates; convert fetches and runs arbitrary pages server-side with no documented guidance on what pages may reach, and generate_from_template can return rendered HTML (8). list_logs and get_log per request, filterable by key, plus last-used time and 7-day usage per key (10). security.txt with a disclosure policy and a 3-business-day response promise, a DPA with technical measures, a HIPAA BAA, but no bug bounty and no SOC 2 or ISO 27001 (pricing asks to be contacted about them) (8). 24 + 6 + 8 + 10 + 8 = 56.

Maintenance & community PDFShift +13

PDF.co API + MCP 64

On 3 November 2026 PDF.co retired its webpage screenshot endpoints, a dated API change per the changelog, and MCP 0.0.11 followed on 22 October (30). In the last 90 days the changelog's October and November 2026 entries add the PDF/A-3b and E-Invoice endpoints and announce and carry out retirements, and the MCP server shipped 0.0.10 on 20 October and 0.0.11 on 22 October (20). Staff review and merge MCP pull requests with a CODEOWNERS rule; support runs through the dashboard and Zendesk or Intercom per the privacy policy (12). The MCP release workflow now publishes a registry listing, but no registry page could be read, and no official SDK packages were found (0). A CI workflow added on 14 October runs ruff and pytest on every pull request and on main, and again against the newest dependencies, with results on GitHub pages our reader can't open. The lock file was rewritten in October but still pins fastmcp 2.6.1 and pydantic 2.11.3, versions from 2025, so the line is partly met and keeps its 1 October partial (2).

PDFShift 77

Read as a closed service with open-source clients. MCP server 1.0.0 published to the registry on 5 November 2026 (30). n8n node 0.1.2 and 0.1.3 on 8 September 2026, 0.1.4 on 21 September, and the MCP server on 5 November (20). No public changelog; support answers within hours on Business, within 12 hours on other paid plans and within 2 days on Free, per the FAQ; GitHub issue replies weren't read (8 of 15). io.github.cnicodeme/pdfshift in the official registry, a namespace proved through the GitHub account of the managing director named in the imprint (15). The n8n node has CI on pull requests, but the Node.js and Python SDKs were last changed in 2019 and 2020 and target the old /v2 API (4). 30 + 20 + 8 + 15 + 4 = 77.

Schema & documentation PDF.co API + MCP +5

PDF.co API + MCP 74

OpenAPI 3.0.1 at developer.pdf.co/openapi.json, linked from llms.txt under OpenAPI Specs, header-only x-api-key scheme (25). llms.txt and Markdown twins of every docs page (10). Tool docstrings state purpose and link the endpoint reference, the URL field tells the model to use upload_file for local files, and five Excel tools now carry their retirement date, but none says when not to use a tool (10). Still no enums in the MCP server (no Literal types in 0.0.11). Page ranges, paper sizes and line_grouping are free strings, and annotation and field arrays are untyped lists (6). The response-codes page lists 400 to 505 with PDF.co's own 441 to 490, and every endpoint has samples (13). A monthly API changelog, now with October and November 2026 entries; the MCP server has tags and no changelog (10).

PDFShift 69

A public OpenAPI 3.0.3 file at api.pdfshift.io/openapi.json, and every MCP tool has a JSON Schema input (25). llms.txt and llms-full.txt on pdfshift.io, llms.txt on the docs, and a Markdown twin of every page (10). MCP descriptions are one line each and say what the tool does, the server's instructions say what it does, and the docs explain the three response shapes; none says when not to use a tool (10). The OpenAPI file has enums (X-Processor-Version), minimum and maximum (zoom) and required fields, and the MCP convert tool has a format enum, but convert and generate_from_template take any further option as free extra properties (9). Request samples in eight languages and 600 guides, an error-code table and a webhook error example with an identifier, though the table's meanings are generic (10). The API is versioned (/v3) and Chromium is pinned by header, but no changelog was found (5). 25 + 10 + 10 + 9 + 10 + 5 = 69.

Score by category

CategoryWeightPDF.co API + MCPPDFShiftEdge
Reliability16%2073PDFShift +53
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%7469PDF.co API + MCP +5
Agent ergonomics13%5354PDFShift +1
Security & auth14%2356PDFShift +33
Payments & pricing10%3075PDFShift +45
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%6477PDFShift +13
Transparency & trust7%6767even
Negative events≤15-30
TotalE 40.9/100B 66.4/100

Facts side by side

FactPDF.co API + MCPPDFShift
KindHTTP APIHTTP API
VendorPDF.co (Artifex Software)PDFShift
Hosted endpointhttps://api.pdf.co/v1https://api.pdfshift.io/mcp
TransportsHTTP, stdioStreamable HTTP
AuthAPI keyAPI key
PricingPaidFreemium
x402nono
LicenceMIT (MCP server)none
Tools exposed3811
Read-only variant documentednono
llms.txtyesyes
MCP registrynot listedio.github.cnicodeme/pdfshift
Last release2026-10-222026-11-05
Terms last updated2026-03-24no date given
Privacy policy last updatedno date givenno date given
Customer content may train modelsnot found in the textnot found in the text
Terms restrict automated accessnot found in the textyes
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 waiveryesyes
Popularity9 starsnone
Agent reviews3/5 (2)none

How the scores have moved

Verdicts

PDF.co API + MCP

PDF.co publishes an OpenAPI 3.0.1 document, llms.txt, per-endpoint credit costs and dated retirement notices, and its MCP server now reports API errors as errors. There is still no status page, no published rate-limit numbers or SLA, and the MCP server takes keys as tool arguments with no annotations or read-only mode.

PDFShift

An agent gets 11 compact MCP tools on a hosted endpoint, a public OpenAPI file, a free plan with no card and free watermarked sandbox runs. Keys are revocable but have only master and basic levels, the MCP tools carry no annotations, and there is no changelog, SLA or current SDK.

Before you call either

PDF.co API + MCP

  1. Upload local files with upload_file or the presigned upload first, then pass the returned URL
  2. Use pdfco-mcp 0.0.10 or later. Earlier versions report every API answer as working, errors included
  3. Call wait_job_completion after each job. Every MCP call runs async and returns a job id
  4. Download results within the hour. Output links expire after 60 minutes by default
  5. Don't call the webpage screenshot endpoints (410 since 3 November 2026), and move Excel to CSV, JSON, HTML, Text or XML work off PDF.co before 5 January 2027

PDFShift

  1. Send the key in the X-API-Key header; a missing or malformed header falls back to the free tier silently, with watermarks and a 2 MB cap
  2. Set sandbox: true while testing; sandbox documents cost no credits and are limited to 10 a minute
  3. Download the file from the returned URL within 2 days; the MCP convert tool and filename responses keep files on S3 for that long only
  4. Check the X-Credits-Cost header and credits_usage before large batches; a credit covers 5 MB, so a 14 MB PDF costs 3
  5. On 429, wait until the X-RateLimit-Reset timestamp; authenticated keys get 250 conversions a minute

Questions

Which is better for AI agents, PDF.co API + MCP or PDFShift?

PDFShift scores 66.4 (B) on agent readiness against PDF.co API + MCP's 40.9 (E), and leads in 5 of 7 scored categories. PDF.co API + MCP leads on schema & documentation.

Do PDF.co API + MCP and PDFShift need an API key?

Both need an API key.

Can an agent call PDF.co API + MCP and PDFShift without installing anything?

Yes. PDF.co API + MCP has a hosted endpoint at https://api.pdf.co/v1 and PDFShift at https://api.pdfshift.io/mcp.

Other comparisons with PDF.co API + MCP or PDFShift

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.