Pipecat Context Hub
Local-first MCP server providing fresh Pipecat docs and examples context for Claude Code, Cursor, VS Code, and Zed.
Quick links: Client Setup | MCP Tools | Version-Aware Queries | Troubleshooting | Environment Variables | Report an Issue
Platform support: macOS and Linux. On Windows the package installs and the CLI help runs, but
refresh,serveand semantic queries currently fail closed (see the Platform contract). Windows improvements are planned: dev plan.
What It Does
When your AI coding assistant needs Pipecat context, it calls MCP tools exposed by this server. The server queries a local index (ChromaDB + SQLite FTS5) and returns relevant documentation, code examples, and API source — all with source citations.
IDE/Agent ←stdio→ pipecat-context-hub serve ←→ Local index (~/.pipecat-context-hub/)
Install
The quickest path is to co-install with the Pipecat CLI, which then hosts every
command below under pipecat context-hub and can wire up your coding agent for you:
uv tool install "pipecat-ai[cli]" --with pipecat-ai-context-hub
pipecat context-hub install # register the MCP server, then build the index
Or run the hub on its own with uv — install it
first if you don't already have it:
curl -LsSf https://astral.sh/uv/install.sh | sh # macOS/Linux
# or: powershell -c "irm https://astral.sh/uv/install.ps1 | iex" # Windows
Once uv is on PATH, uvx fetches and runs the hub on demand; the first
invocation downloads the package and local models (allow a few minutes).
Naming: the PyPI package is
pipecat-ai-context-hub(official pipecat packages arepipecat-ai*); the command and MCP server name arepipecat-context-hub. Both spellings of the command resolve once installed.
The optional skills-only Pipecat plugin
provides Setup, grounded Explore, local Build and approved Cloud Deploy workflows.
It uses pipecat context-hub or standalone pipecat-context-hub through native
execution, with no bundled MCP connection. Plugin installation supplies skills
and assets; ask "Set up Pipecat Context Hub" to check prerequisites and perform
only authorised missing-runtime and initial-index setup. Full-plugin setup also
covers required Pipecat/Cloud CLIs and Cloud account signup/login. Healthy
installations/indexes are reused, and existing standalone MCP clients are unchanged.
Clean local/cloud onboarding and autonomous activation of the CLI-first release
remain unqualified. The earlier stdio experiment's local build and approved READY
Cloud evidence remain dated in the evaluation report; they do not
prove fresh-user, provider/session or ChatGPT Work support for this release.
Platform contract
The persisted Chroma/SQLite index and semantic retrieval path currently require
POSIX descriptor-relative, no-reparse filesystem operations. On Windows the
package and path/ingest helpers remain installable, but refresh, serve and
semantic queries fail closed with an index-safety error rather than using an
unsafe pathname-only fallback. Do not treat the Windows CI package smoke as
index support; use a POSIX host for a populated local index.
Inside the Pipecat CLI
Installing this package alongside pipecat-ai[cli] mounts it as pipecat context-hub,
with pipecat ch as a shorter alias (hidden from pipecat --help, but equivalent).
Discovery is dynamic, so this works with a Pipecat CLI that is already
installed — no upgrade needed on that side.
pipecat context-hub --help # same commands as below
pipecat context-hub refresh
pipecat context-hub check-deprecation PipelineTask
pipecat context-hub <command> and pipecat-context-hub <command> are equivalent —
same parsing, same JSON, same exit codes. Point MCP clients at this package
directly rather than at pipecat context-hub serve, though, so starting the server does
not load the Pipecat CLI first. pipecat context-hub install works out the right command
for your setup — note that a --with co-install exposes only the Pipecat CLI's
own scripts, so pipecat-context-hub is importable but not on PATH.
Populate the Local Index
Before the server can answer queries, build the local index:
# First-time setup (downloads docs, clones repos, computes embeddings)
uvx pipecat-ai-context-hub refresh
# Force full re-ingest (ignores cached state)
uvx pipecat-ai-context-hub refresh --force
# Recover from an unhealthy local index
uvx pipecat-ai-context-hub refresh --force --reset-index
Tip:
refreshreads Pipecat's machine-readablescripts/deprecations/deprecations.json(and, when present,scripts/deprecations/removals.json) from the indexed framework checkout.check_deprecationtherefore does not require GitHub CLI authentication.
Start the Server
Run refresh at least once first (see above). serve exits with code 2
if the index is empty or cannot be opened — it will not start against an
unusable index, since MCP clients would otherwise hang on zero-hit queries.
uvx pipecat-ai-context-hub serve
Client Setup
install does this for you where the client ships a CLI:
pipecat-context-hub install # every detected client, then refresh
pipecat-context-hub install --client codex # just one
pipecat-context-hub install --print-config # show the config, change nothing
Claude Code and Codex are registered through their own CLI, so they own the edit
to their config file. Cursor, VS Code, and Zed are configured by hand, so
install prints the exact JSON and where it goes rather than writing to a file
it does not own. MCP servers are read at session start — restart your agent
afterwards.
Claude Code registrations default to user scope; Codex has no scope concept. Scope
rules, a manual config example and the CLAUDE.md snippet that makes your agent prefer
these tools are in the client setup guides. To set a client up
manually, point its MCP config at uvx pipecat-ai-context-hub serve; for a persistent
install and how serve exits when the client goes away, see
server lifecycle. Per-client guides:
| Client | Setup Guide |
|---|---|
| Claude Code | docs/setup/claude-code.md |
| Cursor | docs/setup/cursor.md |
| VS Code | docs/setup/vscode.md |
| Zed | docs/setup/zed.md |
MCP Tools
| Tool | Use when... |
|---|---|
search_docs |
"How do I ...?" — conceptual questions, guides, configuration |
get_doc |
Retrieve a specific doc page by ID or path (e.g. /guides/learn/transports) |
search_examples |
"Show me an example of ..." — find working code by task or component |
get_example |
Retrieve full source files for a specific example |
search_api |
Class definitions, method signatures, frame types, inheritance |
get_code_snippet |
Get targeted code by symbol name, intent, or file path + line range |
check_deprecation |
Verify whether a pipecat import path is deprecated or removed (supports --at-version for lifecycle at a specific version) |
get_hub_status |
Index health, reranker runtime state, record counts, framework version, commit SHAs, repo coverage (indexed_repos, configured_no_records, catalog_groups_not_indexed) |
All search results include an EvidenceReport with confidence scores, source-grounded facts, unresolved questions, and suggested follow-up queries.
A search response may also carry a top-level corpus_hint when the query
names a repo catalog group whose repos are not indexed: it
lists the missing repos and the repos add / refresh commands that would add
them. An alias matches whole words, including the words of a CamelCase or
snake_case identifier (TavusVideoService, aws_bedrock). The text checked is
query for search_docs and search_examples; query, class_name and
module for search_api; and symbol, intent, class_name and module for
get_code_snippet. The key is absent when no hint applies, and the hint does
not depend on result confidence. The one-shot CLI prints its generic stderr
poor-results hint only when there is no corpus_hint. serve reads configured and tainted repos
once at startup, so restart it after repos add / repos remove.
Filters
search_examples supports filters to narrow results:
domain—"backend"(Python),"frontend"(JS/TS),"config","infra"language—"python","typescript"repo— filter by GitHub repo slugtags— filter by capability tags
search_api supports filters for framework internals:
module— module path prefix (e.g."pipecat.services")class_name— class name prefix (e.g."DailyTransport")chunk_type—"method","function","class_overview","module_overview","type_definition"yields— methods that yield a specific frame typecalls— methods that call a specific method
Multi-Concept Queries
Use + or & to search for multiple concepts at once:
search_docs("TTS + STT")
search_examples("idle timeout + function calling + Gemini")
search_api("BaseTransport + WebSocketTransport")
Each concept is searched independently and results are interleaved for balanced coverage.
Version-Aware Queries
Pass pipecat_version to search_examples, search_api or get_code_snippet to get
results scored for compatibility with your project's pipecat version; results carry
version_compatibility. Details, filters and examples:
Version-Aware Queries.
Index Metadata Contract
External tooling can read how old the index is and which pipecat version it covers
straight from the index_metadata table in <data dir>/metadata.db, a versioned,
read-only contract. Keys, compatibility rules and a copy-paste example:
Index Metadata Contract.
Environment Variables
Every setting is a PIPECAT_HUB_* variable, resolved from real environment variables, a project .env, then ~/.config/pipecat-context-hub/config.toml.
The full variable table and precedence rules are in Configuration; curated repo bundles you can copy into a .env are in .env.example.
Data Sources
The default index includes:
- Pipecat documentation —
docs.pipecat.ai(200+ pages) - Pipecat framework —
pipecat-ai/pipecat(Python AST-indexed: classes, methods, imports, call graphs; includes Pipecat Flows aspipecat.flowsplus itsexamples/flows/) - Pipecat examples —
pipecat-ai/pipecat-examples(project-level code examples) - Daily Python SDK —
daily-co/daily-python(.pyistubs + RST type definitions) - TypeScript SDKs —
pipecat-client-web,pipecat-client-web-transports,pipecat-client-react-native-transports,pipecat-ui,pipecat-prebuilt(tree-sitter-indexed)
To add repos or pin the framework version, see Adding repos. Only Python, TypeScript, and RST are parsed.
The repos command edits PIPECAT_HUB_EXTRA_REPOS in config.toml for you
(rules in Managing repos from the CLI);
run refresh afterwards to index the change:
pipecat-context-hub repos list [--groups] # effective repos (default / catalog / unlisted)
pipecat-context-hub repos add <owner/repo> # also --group NAME, --unlisted, --config-only
pipecat-context-hub repos remove <owner/repo> # same options as add
Maintainers keep the repo catalog current with two
network-only recipes, never run in PR CI: just catalog-health (private, missing,
archived or renamed entries; run weekly by catalog-health.yml) and
just catalog-curate (report-only per-repo stats and vendor candidates).
Security
- Threat model: docs/security/threat-model.md
- Vulnerability reporting: SECURITY.md
- Upstream denylisting:
PIPECAT_HUB_TAINTED_REPOSandPIPECAT_HUB_TAINTED_REFS
Troubleshooting
- Empty results — run
uvx pipecat-ai-context-hub refreshto populate the index - Stale results — run
uvx pipecat-ai-context-hub refresh --forceto re-ingest from latest upstream - Anything else (index corruption and
--reset-index,--prune, low disk, a busy refresh lock,serveexits, Windows, config debugging): see Troubleshooting.
Contributing
See CONTRIBUTING.md for architecture, development workflow, benchmarking, and project structure.
Metadata
Release files for pipecat-ai-context-hub 0.9.1
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| pipecat_ai_context_hub-0.9.1.tar.gz | 1.9 MB | Without provenance | Details for pipecat_ai_context_hub-0.9.1.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| pipecat_ai_context_hub-0.9.1-py3-none-any.whl | Python 3 | none | any | Details for pipecat_ai_context_hub-0.9.1-py3-none-any.whl |
Total release size: 2.2 MB
Release files / pipecat_ai_context_hub-0.9.1.tar.gz
| Download URL | pipecat_ai_context_hub-0.9.1.tar.gz |
|---|---|
| Size | 1.9 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
ed71b4c0ac07168d0fe6c1780213c767affe363bf2ee7b3ebb117da8b7cf8e7d
|
|
BLAKE2b-256 checksum How to use checksums |
939a8e9316e12826848249793686cf556f3c9bbfb0fe170f6f5b9101c6b8ec8f
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
uv/0.9.18 {"installer":{"name":"uv","version":"0.9.18","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}
|
Release files / pipecat_ai_context_hub-0.9.1-py3-none-any.whl
| Download URL | pipecat_ai_context_hub-0.9.1-py3-none-any.whl |
|---|---|
| Size | 326.7 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
dd09dd47818e550ad4242d41c59e71b74dae575fefd97691337ffddd7584049f
|
|
BLAKE2b-256 checksum How to use checksums |
eefb0db8536ecd9ff38e95718a0771d92f0542be4320c5d01535809343dfa256
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
uv/0.9.18 {"installer":{"name":"uv","version":"0.9.18","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}
|
Security for pipecat-ai-context-hub 0.9.1
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.