Pinecone
Long-term Memory for AI. The Pinecone vector database makes it easy to build high-performance vector search applications. Developer-friendly, fully managed, and easily scalable without infrastructure hassles.
Verdict
Common use cases
- Spin up test indexes for prototype features
- Query production embeddings during incident triage
- Create backups before schema migrations
- Monitor index stats and replica health
- Restore namespaces from disaster recovery snapshots
Integration
- Vendor
- Pinecone
- Category
- other
- Auth
- API_KEY
- Tools
- 30
- Composio slug
pinecone
Tools
- Cancel Bulk Import
Tool to cancel a bulk import operation in Pinecone. Use when you need to stop an ongoing import operation that is not yet finished.
- Configure Index
Tool to configure an existing Pinecone index, including pod type, replicas, deletion protection, and tags. Use when you need to scale an index vertically or horizontally, enable/disable deletion protection, or update tags. The change is asy
- Create Backup
Tool to create a backup of a Pinecone index for disaster recovery and version control. Use when you need to preserve the current state of an index including vectors, metadata, and configuration.
- Create Index
Tool to create a Pinecone index with specified configuration. Use when you need to initialize a new vector database index for storing and querying embeddings.
- Create Index from Backup
Tool to create an index from a backup. Use when you need to restore or duplicate index data from a previously saved backup.
- Create Index with Embedding Model
Tool to create a Pinecone index with integrated embedding model for automatic vectorization. Use when you need to set up a new index that automatically converts text to vectors using a pre-configured embedding model.
- Create Namespace
Tool to create a namespace within a serverless Pinecone index. Use when you need to organize vectors into isolated partitions.
- Delete Indexdestructive
Tool to permanently delete a Pinecone index. Use when you need to remove an index from your project. Note: Deletion protection and pending collections can prevent deletion.
- Delete Namespacedestructive
Tool to permanently delete a namespace from a serverless index. Use when you need to remove an entire namespace and all its data. This operation is irreversible and only supported on serverless indexes.
- Describe Backup
Tool to retrieve detailed information about a specific backup. Use when you need to check backup status, configuration, or metadata.
- Describe Bulk Import
Tool to describe a specific bulk import operation in Pinecone. Use when you need to retrieve detailed information about an import's status, progress, timing, and any errors.
- Describe Index Stats
Tool to get index statistics including vector count per namespace, dimensions, and fullness. Use when you need to understand the contents and status of an index.
- Describe Restore Job
Tool to get detailed information about a specific restore job in Pinecone. Use when you need to check the status, progress, or metadata of a restore operation.
- Generate Embeddings
Tool to generate vector embeddings for input text using Pinecone's hosted embedding models. Use when you need to convert text into vector representations for semantic search or similarity matching.
- Get Model Information
Tool to retrieve detailed information about a specific model hosted by Pinecone. Use when you need to understand model capabilities for embedding and reranking operations.
- List Available Models
Tool to list all available embedding and reranking models hosted by Pinecone. Use when you need to discover available models or filter by model type (embed/rerank) or vector type (dense/sparse).
- List Bulk Imports
Tool to list all recent and ongoing bulk import operations in Pinecone. Use when you need to monitor or track the status of data import jobs. Supports pagination with a default limit of 100 imports per page.
- List Collections
Tool to list all collections in a Pinecone project (pod-based indexes only). Use when you need to view available collections.
- List Index Backups
Tool to list all backups for a specific Pinecone index. Use when you need to view available backups for an index. Supports pagination via limit and paginationToken parameters.
- List Indexes
Tool to list all indexes in a Pinecone project. Use when you need to retrieve all indexes with their configurations and status information.
- List Namespaces
Tool to list all namespaces in a serverless Pinecone index. Use when you need to discover available namespaces for data organization. Returns up to 100 namespaces by default with pagination support.
- List Project Backups
Tool to list all backups for indexes in a Pinecone project. Use when you need to retrieve backup information across all project indexes. Supports pagination with limit and paginationToken parameters.
- List Restore Jobs
Tool to list all restore jobs for a project with pagination support. Use when you need to view the status of restore operations or track restore progress.
- List Vectors
Tool to list vector IDs in a Pinecone serverless index. Use when you need to browse or retrieve vector identifiers from a namespace. Supports filtering by prefix and pagination for large result sets.
- Query Vectors
Tool to perform semantic search within a Pinecone index using a query vector. Retrieves IDs and similarity scores of the most similar items, ordered from most to least similar. Either vector or id parameter must be provided.
- Rerank Documents
Tool to rerank documents by semantic relevance to a query. Use when you need to order retrieved documents by their semantic relevance to a user's search query using Pinecone's hosted reranking models.
- Search Records in Namespace
Tool to search records within a Pinecone namespace using text, vector, or ID query. Use when you need to find similar records based on embeddings or record IDs. Results can optionally be reranked for relevance.
- Start Bulk Import
Tool to start an asynchronous bulk import of vectors from object storage (S3, GCS, or Azure Blob Storage) into a Pinecone index. Use when you need to import large volumes of vectors from external storage. Returns an import ID to track the o
- Update Vector
Tool to update a vector in Pinecone by ID. Use to overwrite vector values and/or metadata. Supports bulk updates via metadata filters.
- Upsert Records to Namespace
Tool to upsert text records into a Pinecone namespace. Use when you need to add or update records with automatic text-to-vector conversion.
Setup
Setup guide
- 11. In Switchy, open Settings → Integrations → Browse MCP Servers and select Pinecone. 2. Click Connect and paste your Pinecone API key when prompted (find it in the Pinecone console under API Keys — use a key with write permissions if you plan to create or modify indexes). 3. Switchy validates the key and lists your available projects; select the one you want to work with. 4. Open any Space and type '@Pinecone list my indexes' to confirm the connection works. 5. If you see your index names and stats, you're ready; if you get an auth error, double-check the key has project-level access, not just read-only scope. 6. To invoke a tool, @mention Pinecone in a message and describe what you need — for example, '@Pinecone create a new index called customer-embeddings with 1536 dimensions' or '@Pinecone query the support-docs index for the 5 nearest neighbors to this embedding'.
What teammates see: by default, memories from Pinecone are scoped to the Space (PROJECT visibility) - you can mark any memory PRIVATE or share it ORG-wide.
Works well with
Top models
Compatibility data appears once enough Spaces have used this MCP together with a given model.
How Switchy teams use it
Starter prompts
Create Test Index
@Pinecone create a serverless index named 'qa-test-embeddings' with 1536 dimensions, cosine similarity, and the us-east-1 cloud region.Open in a Space →
Check Index Health
@Pinecone show me the current stats for the 'customer-support' index including total vector count, storage size, and replica configuration.Open in a Space →
Backup Before Migration
@Pinecone create a backup of the 'product-catalog' index and name it 'pre-migration-snapshot-2025-01-15'.Open in a Space →
Query Nearest Neighbors
@Pinecone query the 'docs-embeddings' index for the 10 nearest neighbors to vector [0.23, -0.45, 0.67, ...] and return their metadata.Open in a Space →
Organize with Namespaces
@Pinecone create a namespace called 'team-sales' in the 'company-knowledge' index for storing sales-specific document embeddings.Open in a Space →
Example outputs
Illustrative - representative of the model's voice and quality, not literal recordings.
@pinecone search our product_embeddings index for vectors similar to 'wireless headphones with noise cancellation' and show the top 5 matches with metadata
Searched the product_embeddings index and found 5 similar vectors. Top match (score: 0.94): Product ID 'WH-1000XM5' with metadata {category: 'audio', price: 399, features: ['ANC', 'wireless', '30hr battery']}. Other matches include similar over-ear and in-ear models with scores ranging from 0.89 to 0.91. All results include product names, categories, and pricing metadata for further filtering.
This example demonstrates semantic search against stored embeddings. Requires an existing index with pre-populated vectors and metadata. Query performance depends on index size and pod/serverless configuration. The similarity scores are cosine distance by default but can be configured per index.
@pinecone create a new serverless index called customer_support_docs with 1536 dimensions using cosine similarity, then create a namespace for each product line
Created serverless index 'customer_support_docs' with 1536 dimensions and cosine similarity metric. Index is provisioned in us-east-1 and ready for upserts. Created three namespaces: 'product-line-a', 'product-line-b', and 'product-line-c'. Each namespace can now store vectors independently while sharing the same index infrastructure. Total setup time: ~45 seconds for index initialization.
This example shows index creation and namespace organization. Requires API key with write permissions. Serverless indexes scale automatically but have cold-start latency for first queries. Dimension count must match your embedding model (1536 for OpenAI's text-embedding-ada-002). Namespaces enable multi-tenant data isolation within a single index.
@pinecone fetch usage stats for all my indexes, then recommend which ones should be scaled up based on query volume and which could be consolidated to reduce costs
Analyzed 4 active indexes across your project. Index 'production_search' shows 2.3M queries/day with p95 latency at 180ms — recommend scaling from 2 to 4 replicas for better performance. Index 'staging_test' has only 400 queries/week and could be deleted or merged into a shared development index. Indexes 'legacy_v1' and 'legacy_v2' have overlapping namespaces and similar schemas — consolidating them would save ~$140/month while maintaining query performance.
This example combines Pinecone's monitoring data with AI reasoning for infrastructure optimization. Requires read access to all indexes and their metrics. Recommendations depend on current pricing tier and usage patterns. Always test consolidated indexes in staging before production changes — namespace migrations require careful planning to avoid downtime.
Use-case deep-dives
When Pinecone makes sense for customer support lookup
A 6-person support team fields 200 tickets a week and needs instant semantic search across 10,000 help articles, past tickets, and product docs. Pinecone wins here if your team already has embeddings infrastructure or uses a platform that generates vectors automatically. The Create Index with Embedding Model tool handles text-to-vector conversion, so agents can query in plain language without preprocessing. The namespace feature lets you partition customer data from internal docs, which matters for compliance. The trade-off: if you're starting from scratch and don't have an embedding pipeline, you'll spend a week on setup before the first search works. If your knowledge base is under 2,000 articles and updates rarely, a simpler full-text search tool costs less and ships faster. Pinecone pays off when search quality directly impacts resolution time and your content changes daily.
Why Pinecone fits early-stage recommendation builds
A 3-person product team at a marketplace startup is prototyping a recommendation engine to surface similar listings based on user behavior. They have 50,000 product embeddings from an external model and need to test different similarity thresholds before committing to infrastructure. Pinecone's serverless indexes let them spin up, configure, and delete test environments without provisioning servers—the Configure Index and Delete Index tools make iteration cheap. The Create Backup tool is critical here: before each experiment, they snapshot the index so they can roll back if a configuration tanks performance. The boundary: once you hit production scale with millions of vectors and sub-50ms latency requirements, you'll need to evaluate pod-based indexes or self-hosted alternatives. For prototyping and MVPs under 100k vectors, Pinecone's API-key simplicity and backup tooling beat managing your own vector database.
When namespace isolation justifies Pinecone for SaaS
A 10-person SaaS company building a document analysis platform serves 200 customers, each with their own corpus of contracts and reports. They need semantic search that keeps customer data strictly isolated without spinning up 200 separate databases. Pinecone's Create Namespace tool solves this: one serverless index, 200 namespaces, each customer's vectors partitioned and queryable independently. The API-key auth model means each customer request can scope to their namespace programmatically, and the Configure Index tool lets them scale replicas as query load grows. The catch: if your customers expect to export or migrate their vector data easily, Pinecone's proprietary format adds friction. And if you're under 20 tenants, the namespace overhead isn't worth it—just use separate indexes. Pinecone makes sense when tenant count is high, data isolation is non-negotiable, and you'd rather pay for managed infrastructure than hire a vector-ops engineer.
Frequently asked
What does the Pinecone MCP do in Switchy?
It lets your team create, configure, and manage Pinecone vector indexes directly from Switchy's AI workspace. You can create indexes with embedding models, organize vectors into namespaces, run bulk imports, create backups for disaster recovery, and delete indexes when needed. All 30 tools operate on your Pinecone project using your API key.
Do I need admin access to connect Pinecone?
You need a Pinecone API key with write permissions to your project. Pinecone uses API key authentication, not OAuth, so whoever connects it must have key-generation rights in your Pinecone console. Read-only keys won't work because most tools create or modify indexes. One team member connects it; everyone in the Switchy workspace can use it.
Can the Pinecone MCP query vectors or just manage indexes?
This MCP focuses on index management — creating indexes, configuring pod types and replicas, setting up namespaces, running backups, and bulk imports. It doesn't include query or upsert tools for working with individual vectors. For querying embeddings or inserting data, you'd still use Pinecone's SDK or REST API directly in your application code.
Why use this instead of the Pinecone console or API?
The console requires context-switching and manual clicks. The API requires writing code. This MCP lets your team manage indexes conversationally — "create a backup of the prod index", "scale replicas to three" — without leaving Switchy. Useful for ops tasks, disaster recovery, and letting non-engineers provision indexes without touching infrastructure code.
Who on the team should connect the Pinecone integration?
Whoever owns your Pinecone project and can generate API keys. Usually an engineer or DevOps lead. Once connected, anyone in the Switchy workspace can trigger index operations, so consider whether you want junior team members creating or deleting production indexes. Pinecone's deletion protection setting can prevent accidental drops.