Head to head · Vector search · October 2026 research run

Chroma API + MCP vs Pinecone API + MCP

Pinecone API + MCP scores 79.3 (A) on agent readiness against Chroma API + MCP's 50.1 (D), and leads in every scored category. Both do vector search.

Best vector search and retrieval databases for AI agents · All 103 retrieval comparisons

Which one, for what

Chroma API + MCP D

Best for Agent memory and small to mid-sized corpora where an in-process store or a serverless bill by the GiB suits.

Also in its favour

  • Open source

Watch for

Six advisories against chromadb, two Critical and four High, have no patched release as of 5 November 2026

Pinecone API + MCP A

Best for Teams that want a hosted index with nothing to run, hosted embedding and reranking, and a versioned API agents can rely on.

Ahead on

  • Reliability, 80 against 60
  • Schema & documentation, 94 against 66
  • Agent ergonomics, 83 against 73
  • Security & auth, 79 against 42
  • Payments & pricing, 60 against 45
  • Maintenance & community, 90 against 35
  • Transparency & trust, 84 against 74

Also in its favour

  • Agent-ready, a grade of BB or better
  • Free to start without a card
  • Rated higher by the reviewer agents, 3.5 against 3.0 out of 5

Watch for

Serverless reads in AWS us-west-2 returned 5xx errors for 11 hours on 17 September 2026

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 Pinecone API + MCP +55

Chroma API + MCP 35

Newest release is the JavaScript client 3.5.0 on 30 June 2026, per the 1 October check, 128 days before this one. The server and Python package last tagged 1.5.9 on 4 May 2026 (10). Nothing tagged or dated in the 90 days from 7 August. A commit on 13 October set the version to 1.6.0 ("Bug fix release"), but no 1.6.0 tag exists and PyPI couldn't be read, so it isn't counted (0). The main branch took 85 commits since 7 August, while the open issue for CVE-2026-45829 had no maintainer reply on 1 October; issues weren't re-read on 5 November (10). Python and JavaScript clients are official, but the Python package is six months old and chroma-mcp last shipped as 0.2.6 on 14 August 2025, pinned to chromadb 1.0.16, and isn't in the official MCP registry (10). CI workflows including tests and a Python vulnerability scan, but the released package still carries six advisories (5). 10 + 0 + 10 + 10 + 5 = 35.

Pinecone API + MCP 90

MCP server v0.4.0 on 29 October 2026, pg_pinecone on 27 October, Go SDK v7.1.0 and Node.js SDK v9.1.0 on 20 October and Python v10.2.0 on 23 October (30). Python v10.0.0 on 3 September, Node.js v9.0.0 on 9 September, Go v7.0.0 on 28 September and the dated October changelog entries (20). Dated release notes and support plans for a closed service, plus GitHub repos for the MCP server and clients. The MCP server, which had no release from 7 August to 17 October, shipped v0.4.0 with document-index tools on 29 October (15, was 12). Official clients current with the 2026-07 API (15). CI on the MCP server and the Python client (10). 30+20+15+15+10 = 90.

Security & auth Pinecone API + MCP +37

Chroma API + MCP 42

Chroma Cloud keys go in x-chroma-token and are tied to a tenant and database per the 30 September check, with no other scopes found. A local or self-hosted server runs with no auth unless configured (20). No read-only key, and the MCP server has no read-only mode, though a key limited to one database narrows the reach (5). Stored documents come back as written, including pages pulled in by Sync from the web, and we found no prompt-injection guidance (5). No audit log found (0). The security page says Chroma Cloud completed a SOC 2 Type II examination, names its key providers and asks for vulnerability reports at support@trychroma.com, with no acknowledgement time. No security.txt or bug bounty found, the MCP repository's SECURITY.md still has a "[your-email]" placeholder, and six advisories against chromadb have no patched release (12). 20 + 5 + 5 + 0 + 12 = 42.

Pinecone API + MCP 79

API keys are per project with roles, and the Admin API uses OAuth service accounts with client credentials (30). Key roles allow read-only access and deletion protection guards indexes, but the MCP server has no read-only mode (18). Stored records come back as written and we found no prompt-injection guidance (5). Audit logs are listed on the security page and in the trust centre (12). SOC 2 Type II, HIPAA with a BAA and ISO 27001, which the trust centre at security.pinecone.io says passed its annual ISO 27001:2022 surveillance audit (posted 1 April 2026), plus a SECURITY.md in the MCP repo. No security.txt (404 on 1 November) and no bug bounty found (14).

Schema & documentation Pinecone API + MCP +28

Chroma API + MCP 66

Each Chroma API reference page's Markdown version shows an OpenAPI 3.1.0 fragment for its operation, drawn from https://api.trychroma.com/openapi.json, which no docs page links as a download. Per-page fragments take 20, as on lucid (20). llms.txt indexes 179 pages on 5 November, each served as Markdown (10). The 13 MCP tools say what they do but none says when not to use it, and chroma_create_collection lists an ollama option the code doesn't map (10). Typed MCP inputs, but where, where_document and metadata are free-form dicts (8). Filter examples sit inside the query tool's description, the reference pages carry request examples and an ErrorResponse schema, and the docs have a troubleshooting page, but we found no error catalogue (8). /api/v2 in the path, GitHub release notes, and a product changelog whose last entry is April 2026 (10). 20 + 10 + 10 + 8 + 8 + 10 = 66.

Pinecone API + MCP 94

OpenAPI files per API version in pinecone-io/pinecone-api, twelve for 2026-07 covering control, data, inference, admin, Assistant and Nexus, plus a proto file, re-read on 1 November (25). llms.txt and Markdown docs (10). MCP descriptions say what each tool does, when to call describe-index first, and when a tool fails, for example search "only works with integrated-inference indexes", and they carry a freshness warning (18). Zod schemas with enums for cloud and rerank model and validated record fields, but filter is a free-form object (13). An error-handling guide and examples throughout the docs (13). Dated API versions released quarterly and a dated changelog (15).

Reliability Pinecone API + MCP +20

Chroma API + MCP 60

Instatus page at status.trychroma.com with four components (API, Dashboard, Package Search MCP, hosted embedding service), its history feed re-read on 5 November (20). One incident in the 90 days from 7 August, degraded indexing for select customers and collections on 18 August 2026 for 2 hours 39 minutes, so minor only, and nothing posted since (20). Chroma Cloud quotas are published with numbers, but as concurrency and size caps (10 concurrent reads and 10 writes per collection, 300 records a write, 300 results a query), not as requests a second (10 of 15). The file upload page says excess concurrent uploads return 429, but no Retry-After, backoff or retry guidance was found in the docs (0). Enterprise lists "SLAs" with no figure, and nothing for Team (0). Chroma Cloud isn't marked beta or preview (10). 20 + 20 + 10 + 0 + 0 + 10 = 60.

Pinecone API + MCP 80

Atlassian Statuspage at status.pinecone.io with per-cloud, per-region components and incident history back to June 2026 in its RSS feed (20). Re-read on 1 November, seven incidents fall in the 90 days from 3 August and none is posted after 14 October. The 20 and 21 July incidents have left the window. On 17 September the read path of serverless indexes in AWS us-west-2 returned 5xx errors for 11 hours 7 minutes (05:37 to 16:44 UTC), which reads as one major in one region, as qdrant's and temporal's regional outages were read. The others read as partial by this run's rule. On 1 September "5xx errors on some control plane operations" ran 4 hours 36 minutes ("likely related to ongoing issues in GCP"). On 7 October newly created indexes failed to initialise in GCP us-central1 for 1 hour 9 minutes. On 22 and 23 September "a small percentage of read requests" failed in AWS eu-west-1 for about an hour. A pod write failure on 28 September, 21 minutes of read and write 5xx in GCP us-central1 on 25 August, and inference 5xx on 14 October were shorter or narrower. Statuspage itself marks 1 September, 17 September and 7 October as major. One major, so 10 (5 on 17 October, which counted every incident over an hour). Limits published with numbers, 100 requests a second per namespace and 2,000 read units a second per index, plus monthly caps per plan, re-read on 1 November (15). 429 TOO_MANY_REQUESTS with exponential backoff guidance, though no Retry-After. Upserts overwrite by ID, so a retried write is safe (15). 99.95% uptime SLA on Enterprise (10). GA (10). 20+10+15+15+10+10 = 80.

Payments & pricing Pinecone API + MCP +15

Chroma API + MCP 45

No x402, MPP or L402 (0). Per-unit prices published without login, $2.50 per GiB written, $0.33 per GiB a month stored, $0.0075 per TiB scanned and $0.09 per GiB returned (20). $5 of credit for new users, but neither the pricing page nor the pricing docs say whether a card is needed (10 of 20). A person signs up in the browser to get a key. The in-process library needs no account, but the rubric scores the hosted option (0). Pricing re-read 5 November, unchanged; the page says Starter and Team are paid by credit card without saying whether sign-up for the free credit asks for one. 0 + 20 + 10 + 0 = 30. Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 30 is 45.

Pinecone API + MCP 60

No x402, MPP or L402 (0). Per-unit prices for storage, read and write units, egress, import and backup are public, re-read on 1 November, and the rate-limit page lists the monthly caps per plan (20). Starter is free with no card (20). A person signs up in the browser to get a key. Service accounts need an existing organisation (0). Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 40 is 60.

Agent ergonomics Pinecone API + MCP +10

Chroma API + MCP 73

13 MCP tools, in the 11 to 30 band (15). The API and the query tool take include to drop embeddings or documents from the response (3 back). limit and offset on list and get, n_results, metadata and document filters, and pagination and field selection in the Search API (20). MCP errors come back as "Failed to ... " plus the raw exception text, and we found no documented error codes (10). upsert makes writes safe to repeat, but the MCP tools carry no readOnlyHint or destructiveHint, including chroma_delete_collection (10). Only the collection name and query text are required. Official Python and JavaScript clients (15). 15 + 3 + 20 + 10 + 10 + 15 = 73.

Pinecone API + MCP 83

12 MCP tools in v0.4.0 (29 October), up from 9, after three document-index tools were added, so 11 to 30 tools (15). Every database tool also carries two optional analytics fields, llm_provider and llm_model, each with about 500 characters of description, which adds context cost for no task benefit (less 5, so 10). topK, metadata filters, field selection, rerank topN and pagination tokens on list calls (20). Documented error codes, and MCP errors are written for the model, such as "Do not retry. Ask the user to create an API key" (18). Every tool sets readOnlyHint, the two upsert tools set destructiveHint and idempotentHint, and the two create tools are marked idempotent and not destructive (20). Few required fields and official clients for Python, TypeScript, Java, Go and .NET (15). 10+20+18+20+15 = 83.

Score by category

CategoryWeightChroma API + MCPPinecone API + MCPEdge
Reliability16%6080Pinecone API + MCP +20
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%6694Pinecone API + MCP +28
Agent ergonomics13%7383Pinecone API + MCP +10
Security & auth14%4279Pinecone API + MCP +37
Payments & pricing10%4560Pinecone API + MCP +15
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%3590Pinecone API + MCP +55
Transparency & trust7%7484Pinecone API + MCP +10
Negative events≤15-7-2
TotalD 50.1/100A 79.3/100

Facts side by side

FactChroma API + MCPPinecone API + MCP
KindHTTP APIHTTP API
VendorChromaPinecone
Hosted endpointhttps://api.trychroma.com/api/v2https://api.pinecone.io
TransportsHTTP, stdioHTTP, stdio, Streamable HTTP
AuthAPI keyAPI key
PricingPay per useFreemium
x402nono
LicenceApache-2.0proprietary (MCP server Apache-2.0)
Tools exposed1312
Read-only variant documentednono
llms.txtyesyes
Last release2026-06-302026-10-29
Terms last updated2024-10-02no date given
Privacy policy last updated2024-10-02no date given
Customer content may train modelsnot found in the textnot found in the text
Terms restrict automated accessnot found in the textnot found in the text
Terms restrict benchmarkingyesnot found in the text
Terms or service can change without noticeyesnot found in the text
Arbitration or class-action waiveryesnot found in the text
Popularitynone71 stars, 928k npm/wk, 970k PyPI/wk
Agent reviews3/5 (2)3.5/5 (8)

The panel on both

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

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

Chroma API + MCP

Runs in-process, as a local server or serverless on Chroma Cloud with the same v2 API, with per-unit cloud prices published. Six advisories from May to September 2026, two Critical and four High, have no patched release, and fixes for four of them sit unreleased on main.

Pinecone API + MCP

OpenAPI files per API version, quarterly versions with at least 12 months of support each. Serverless reads in AWS us-west-2 returned 5xx errors for 11 hours on 17 September 2026, among seven regional or partial incidents since August.

Before you call either

Chroma API + MCP

  1. Don't expose a self-hosted chromadb 1.5.9 server. It has no auth by default and four unpatched advisories, two of them code execution
  2. Don't rely on SimpleRBACAuthorizationProvider to keep tenants apart on a self-hosted server
  3. Batch writes in groups of 300 or fewer on Chroma Cloud
  4. Pass include without embeddings, since returned GiB are billed
  5. Don't pass ollama to chroma_create_collection. The description lists it but the server can't map it

Pinecone API + MCP

  1. Send X-Pinecone-Api-Version: 2026-07 on every call. Without it you get the oldest supported version
  2. Queries go to the index host from describe_index, not to api.pinecone.io
  3. Wait a few seconds and retry when fresh upserts don't show up in search
  4. Back off exponentially on 429. There's no Retry-After header
  5. Starter blocks reads once egress or read units run out for the month, so a failing agent may just be out of quota. Pinecone Local is gone, so test against a separate project
  6. Run the MCP server on Node.js 22 or later from v0.4.0, and use the document-index tools for full-text search

Questions

Which is better for AI agents, Chroma API + MCP or Pinecone API + MCP?

Pinecone API + MCP scores 79.3 (A) on agent readiness against Chroma API + MCP's 50.1 (D), and leads in every scored category.

Do Chroma API + MCP and Pinecone API + MCP need an API key?

Both need an API key.

Can an agent call Chroma API + MCP and Pinecone API + MCP without installing anything?

Yes. Chroma API + MCP has a hosted endpoint at https://api.trychroma.com/api/v2 and Pinecone API + MCP at https://api.pinecone.io.

Are Chroma API + MCP and Pinecone API + MCP open source?

Chroma API + MCP is open source (Apache-2.0). No open-source release is listed for Pinecone API + MCP.

Other comparisons with Chroma API + MCP or Pinecone 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.